Ostratto - Blog | Ostratto #artificial intelligence

AI and the future of work income and inequality

Mon, 09 Nov 2026 10:19:30 +0000

AI could allow society to produce more with less human labour. Whether that translates into more leisure, greater insecurity or a wider gap between rich and poor will depend on how income, ownership and public services change alongside it.

Imagine a business that can deliver the same service with half the human effort. Its owners could reduce prices, increase profits, serve more customers, improve pay or shorten the working week. They could also reduce the number of people they employ.

All of those outcomes begin with the same technological improvement. Their consequences for the people involved are very different.

That is the difficulty at the centre of the debate about AI and the future of work. We can imagine machines doing a growing share of economically valuable activity. It is much harder to explain how everyone will live comfortably if paid employment becomes less necessary.

For most households, work provides the money that buys access to housing, food, energy and everything else. A future in which society needs less labour would therefore require more than better technology. It could require a different relationship between work and the right to a decent standard of living.

There is no reliable date for that future, and no certainty that it will arrive in its most dramatic form. But exploring it helps explain the decisions already facing businesses and governments.

What the evidence tells us about jobs

The claim that AI will eliminate most employment goes much further than the research can currently establish.

The International Labour Organization and NASK estimated in May 2025 that one in four jobs worldwide was in an occupation potentially exposed to generative AI. Their assessment said that changes to jobs were more likely than their complete replacement. Exposure measures the potential for technology to affect tasks; it does not count redundancies or predict how quickly employers will adopt it.

A job usually combines many activities. Producing a report, handling an unusual complaint, taking responsibility for a decision and understanding what a customer actually needs may sit within the same role. Automating one activity does not automatically remove the rest.

There is also evidence that AI can help people work more effectively. A study published in The Quarterly Journal of Economics in 2025 examined 5,172 customer-support agents and found that access to an AI assistant increased issues resolved per hour by 15% on average. Less experienced and lower-skilled workers gained more than their more experienced colleagues. That is evidence from a particular workplace, rather than a prediction for the whole economy.

These findings leave room for several outcomes. Employers might expand services with the same team, employ fewer people, or combine both approaches. Productivity improvements do not determine which choice they make.

Software capability also differs from economical, dependable deployment. Systems need to work with existing processes, handle mistakes and justify their operating costs. Extending automation into construction, maintenance or personal care introduces physical challenges that a more capable chatbot alone does not solve.

Why a richer economy might still leave households worse off

The most important distinction is between producing wealth and distributing purchasing power.

Suppose a fictional company earns £1 million a year and spends £600,000 on wages. Imagine that automation eventually allows it to maintain its output while reducing that wage bill to £300,000. Assume, purely for illustration, that the technology and its operation add £100,000 in annual costs.

The business would then have £200,000 more available before any other changes. That money could support investment, lower prices, higher pay, shorter hours or greater returns to owners. The £300,000 reduction in wages could involve fewer staff, reduced hours or another change in employment.

This example is arithmetic, not a forecast. It shows why a more productive business does not automatically make every affected household better off.

Consumers might gain through lower prices while some workers lose their main source of income. Investors might benefit through higher profits while people without savings have little direct claim on those returns. The economy could grow overall while particular communities become poorer.

The IMF has identified both labour-income and ownership channels through which AI could increase inequality: some workers may benefit more than others, while higher returns to capital could favour people who already own assets. These are risks, rather than an unavoidable outcome.

The distribution of productivity gains can follow several routes:

Four possible futures for work

There is no single economic theory that tells us precisely what comes next. Four broad scenarios help explain the range of possibilities.

Possible future	How people earn a living	What could happen to inequality
AI assists most workers	Wages remain central, with jobs changing around AI	Gaps could narrow where tools spread expertise, or widen where access and bargaining power differ
Productivity supports shorter working weeks	People retain paid work but share some gains through more free time	Benefits depend on whether lower hours preserve adequate income and extend beyond privileged occupations
Ownership becomes more important than employment	Profits, assets and access to platforms matter more, while wages become less secure	Wealth could concentrate if displaced workers have little ownership or protection
Much less paid work is needed	Income comes from a mixture of remaining jobs, public transfers, services and ownership returns	Living standards depend heavily on who controls production and how its benefits are shared

These scenarios could coexist. A consultant might become more productive, an administrator might lose a role, a care worker might remain in high demand and a shareholder might receive higher returns during the same period.

The path between them matters as much as the eventual destination. A society that becomes wealthier over twenty years can still impose severe losses on people who cannot pay their bills during the transition.

How people could live if there were less paid work

If employment became a less reliable way to distribute income, households would need other sources of security. Several approaches could contribute, but each answers a different part of the problem.

Shorter working weeks

One possibility is to distribute some productivity gains as time. If a team can complete its work in fewer hours, an employer could preserve pay while reducing working time. People would still earn wages, but employment would occupy less of their lives.

This is different from cutting someone's hours and income because there is less work available. A shorter working week only represents a shared benefit if households can still afford to live.

It would also be harder to deliver evenly across occupations. Reducing office administration may free time directly; maintaining round-the-clock physical care still requires people to cover shifts unless technology changes the underlying work.

Shorter hours are therefore a possible route towards sharing gains, rather than an automatic consequence of AI adoption.

A guaranteed income

Universal basic income, usually shortened to UBI, would provide a regular payment to everyone within an eligible population without making it conditional on employment.

Its appeal is straightforward: people would retain a financial floor even when their work disappeared, their hours varied or they needed time to retrain. Other designs, such as a guaranteed minimum income, would target support more closely according to income.

But the amount and funding matter enormously. A modest payment might improve security without being sufficient to replace wages. A payment large enough to support most living costs would require a much more substantial funding arrangement.

For scale, an illustrative payment of £1,000 a month to 50 million adults would have a gross annual cost of £600 billion. That is not a proposal or a net budget estimate. Taxes paid back by recipients and any benefits replaced would change the net cost, while protecting disability, housing and other additional needs would affect the design.

Finland's 2017 to 2018 basic-income experiment found better reported wellbeing and economic security, alongside small employment effects whose interpretation was complicated by another policy change. It provides useful evidence about a limited trial among unemployed people, rather than proof that a permanent national scheme could support a highly automated economy.

A credible income proposal must also explain what happens to essential costs. If housing supply remains constrained, additional purchasing power could partly feed into higher rents. That is a reason to consider income and housing together, rather than assume that a cash payment alone resolves insecurity.

Better access to essential services

Households can become more secure through a lower cost of living as well as a higher cash income.

Affordable housing, transport, education, childcare and reliable public services could reduce the amount people need to earn privately. AI might help provide some services more efficiently, although translating administrative savings into better provision would still require investment and competent delivery.

This approach is often discussed as universal basic services. Its practical question is whether people can obtain what they need, rather than simply whether they have received a payment.

Cash and services need not be competing choices. A future with less paid work could require both personal spending power and dependable access to essentials.

A wider share of ownership

Redistribution after profits have been earned is only one way to share the benefits. Another is to spread ownership of the assets generating them.

Employee ownership, profit-sharing, pension investment and publicly owned investment funds are possible mechanisms. A public fund could, in principle, hold investments and use returns to support services or payments to citizens.

None offers free or guaranteed income. Funds need assets, returns fluctuate and governance matters. Employee shares also expose workers to the fortunes of their own employer, which can compound risk if both their job and investment lose value together.

The underlying principle is still relevant: if a greater share of income comes from owning productive assets, access to ownership becomes more important to economic security.

Who would pay for a society with less work

A more automated economy would still have to produce the homes, food, energy and services that people consume. Money transfers cannot replace that productive capacity.

If AI expanded output substantially, society might have more resources available to share. The challenge would be creating institutions that convert that capacity into household security.

Public funding could come from a mixture of taxes, investment returns and changes to existing expenditure. If wages became a smaller share of economic income, governments might need to reconsider how heavily revenue collection depended on employment.

The IMF's 2024 discussion of fiscal responses to AI recommended stronger social protection and consideration of capital-income and corporate taxation. It also cautioned against a specific tax on AI itself, citing risks to investment and difficulties with implementation.

There is no simple answer in taxing every automated task. Governments would need to distinguish productive investment from excessive concentration of economic power, collect revenue effectively and avoid undermining the growth expected to fund better living standards.

International coordination would matter where profits and productive assets crossed borders. A country could experience disruption to local employment without automatically receiving a proportionate share of the gains.

Business incentives would remain important too. Companies need customers with purchasing power, but that does not mean each company has an incentive to preserve employment on society's behalf. A business can cut its own wage bill while relying on customers who earn money elsewhere. Wider economic security therefore cannot depend solely on voluntary decisions by individual employers.

Why cheaper intelligence would not make everything cheap

A common vision of the AI future assumes that dramatically cheaper intellectual work will make life broadly affordable. That could happen in some areas without happening everywhere.

Producing another digital document or software feature is different from supplying another home in a desirable location. Land, building materials, energy networks, physical infrastructure and human attention can remain scarce.

AI could help improve planning, design, logistics or research. Those improvements would not automatically create the physical capacity or political agreement needed to increase supply.

A household might gain access to excellent low-cost software while still struggling with rent. A business might generate marketing material cheaply while paying heavily for distribution, premises or access to customers.

Affordability would depend on which costs fell, whether competition passed savings through, and whether the essentials occupying household budgets became easier to obtain. A comfortable life requires more than abundant digital output.

How inequality could change

AI could affect several kinds of inequality at once, sometimes in opposing directions.

Within a workplace, accessible tools might help less experienced people perform tasks that previously required substantial training. The customer-support study offers one example of that possibility. It does not establish that wages will converge, because pay also depends on labour demand, institutions and bargaining power.

Between households, the effects could move in the opposite direction. Someone with a home, investments and a secure professional position may have several ways to benefit from automation. Someone relying entirely on a vulnerable wage may have much less protection.

The result could be a smaller gap in certain skills alongside a larger gap in wealth and security.

There is also a question of market power. In April 2024, the UK's Competition and Markets Authority warned that control of important inputs and routes to market could allow powerful technology firms to restrict competition in foundation-model markets. Its concern was that businesses and consumers could lose out through reduced choice, higher prices or weaker innovation.

For smaller businesses, that creates an unresolved tension. AI could make capabilities affordable that previously required a large team. It could also increase dependence on suppliers that control models, infrastructure or customer access. Whether small firms gain lasting bargaining power would depend partly on how open and competitive those markets remain.

Geography matters as well. Lower-cost access to expertise could benefit people far from major economic centres. But a community losing a significant source of employment would not necessarily recover just because useful software became available.

Access to tools is one part of opportunity. Time, education, infrastructure, capital and a realistic route to earning income are others.

The transition could be harder than the eventual future

Even if AI ultimately supports higher living standards, the adjustment could be painful.

A person losing a job needs income immediately. Training takes time, available vacancies may be elsewhere, and a new occupation may offer lower pay. “Reskilling” only becomes a meaningful answer when there is accessible training, financial support and work worth moving into.

The transition could also affect how people enter professions. If businesses automate activities traditionally assigned to junior staff, they would need alternative ways to develop experienced employees. That is a risk to plan for, rather than proof that every entry-level role will disappear.

For employers, useful questions include how productivity gains will be shared, how staff will learn, and where human judgement remains necessary. Measuring only output or headcount would miss changes in service quality, workload and the development of future expertise.

For governments, the equivalent questions concern income support, housing, training and the speed at which assistance reaches affected communities. An eventual increase in national output would offer little comfort to someone facing several years of financial instability.

What people would do if employment occupied less of life

Paid work provides more than money. For many people it offers routine, relationships, recognition and a sense of contribution. A future with less employment would need to take those roles seriously.

But useful activity extends well beyond paid jobs. Raising children, caring for relatives, supporting charities, creating art, learning and participating in local communities can all have considerable value without generating a wage.

Greater financial security and more free time could allow more of that activity. It could also allow people to start businesses or pursue projects that cannot immediately support them.

The experience would depend heavily on choice. Working fewer hours with a secure income is very different from being excluded from employment while struggling to afford essentials. Describing both situations as “people not working” hides the most important difference between them.

A successful future would need room for contribution without treating conventional employment as the only measure of a person's worth. It would also need to recognise that preferences differ: some people would welcome more leisure, while others would continue to seek demanding paid work.

The future depends on how the gains are shared

There is a credible possibility that AI will help people accomplish far more with less effort. There is much less certainty about whether it will remove the need for most paid employment, how quickly any transition would happen, or which occupations would be affected most.

The sensible response is to prepare for several outcomes. Businesses can use AI to improve services and productivity while making deliberate choices about staff development, workload and the distribution of gains. Governments can strengthen the systems that help people navigate changes in employment, while considering how income and ownership might evolve if those changes become more profound.

The most optimistic future would combine productive technology with economic security and greater freedom over how people spend their lives. A less favourable one would combine impressive output with concentrated ownership, fragile incomes and limited influence over the systems on which people depend.

The technology could contribute to either. Whether people can live well with less paid work will depend on the arrangements built around it: who owns productive assets, how markets operate, which essentials are affordable, and how people obtain a fair share of what society produces.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

AI Agents and the Future of Business Software

Thu, 08 Oct 2026 09:00:00 +0000

Businesses have spent years investing in software to make work easier, yet a considerable part of the working day still involves operating that software. Employees search for records, update fields, move information between applications and work out which screen contains the answer they need. Even when each application is reasonably good, the person using it often becomes responsible for joining everything together.

We think this is likely to change substantially. Our expectation is that AI agents will become the primary way people interact with software for many routine business tasks. Employees will describe what they want to achieve, while agents retrieve information, use the appropriate tools and carry out work across the systems underneath.

That future will develop unevenly, and there are good reasons for people to retain direct control over certain tasks. Nevertheless, it has practical implications for businesses today. The quality of your data, the structure of your processes and the accessibility of your applications could soon have a much greater influence on how effectively your team works.

From navigating applications to asking for outcomes

Consider what happens when an account manager needs to prepare for a customer meeting. They might open the CRM to review recent conversations, check the accounting system for outstanding invoices, look through support tickets and visit the project management platform to understand how ongoing work is progressing.

The information is already available, but gathering it requires the employee to know where to look and how to interpret what they find. A straightforward request becomes a small research exercise across several applications.

An agent with suitable access could potentially assemble that information from a single instruction:

“Prepare me for tomorrow’s meeting with this customer. Summarise recent activity, outstanding issues and anything we need to discuss.”

The more significant opportunity comes afterwards. Having reviewed the summary, the employee could ask the agent to prepare an agenda, create an internal follow-up task or draft a message to the project manager. With the appropriate tools and permissions, the agent could carry out those actions in the relevant systems.

AI systems can already be designed to use tools and choose steps towards an objective. Anthropic’s guidance on building effective agents describes this distinction between predefined workflows and agents that determine how to proceed as they work. That capability provides a foundation for the change we expect, although dependable business use still requires careful implementation. anthropic.com

The employee’s attention can then move towards the customer and the decisions that need making, with less time spent navigating software to gather the necessary context.

Software becomes the infrastructure behind the work

If agents become the main point of interaction, the applications underneath them will continue to perform essential functions. Accounting software will maintain financial records, the CRM will organise customer relationships, and project management systems will hold tasks, responsibilities and deadlines.

What changes is how often an employee needs to open each application directly.

We can imagine a working day in which someone reviews priorities through an assistant, asks it to investigate an issue and approves a proposed action without visiting every system involved. Some tasks may begin through a conversation, while others start automatically when an agreed condition is met.

A renewal process, for example, could bring together contract dates, current services, account activity and outstanding issues before presenting a proposed renewal for review. The person responsible would still make the commercial decision, but much of the preparation could happen in the background.

This would also change what businesses value when choosing software. The interface would remain relevant, but buyers would have stronger reasons to examine the capabilities underneath it: whether information can be retrieved reliably, whether actions can be performed through integrations, and whether access can be restricted appropriately.

A feature that is available only through a sequence of manual screens may be much harder to incorporate into an agent-led process.

Interfaces will still have a role

We expect businesses to spend less time navigating application interfaces, but that does not mean every task should become a conversation.

A table can be a better way to compare hundreds of records. A project board can make workload easier to understand. A chart can reveal a trend more clearly than several paragraphs of explanation. People will also need ways to inspect evidence, review proposed changes and investigate mistakes.

The likely direction is a combination of agents and interfaces, with the agent bringing forward the view needed for a particular decision. Someone might ask which projects are falling behind and receive an interactive table showing deadlines, blockers and responsible team members. They could examine the details before asking for follow-up actions.

In that environment, interfaces become more focused on helping people understand and control the work. Routine navigation and administration can increasingly happen behind the scenes.

Why MCP matters to this future

For agents to work across business applications, they need a reliable way to access information and perform actions. Model Context Protocol, or MCP, provides an open standard for connecting AI applications to external tools and data sources. An MCP server can expose specific capabilities that a compatible AI application can discover and use. Model Context Protocol

We explored the underlying technology in our earlier article, What is MCP, and what does it mean for businesses?. Its relevance here is that a common connection standard can make business systems more accessible to agents.

For example, a connection might expose tools for finding customer records, retrieving invoice information or creating tasks. The agent can request those operations using defined inputs, while the connected system performs the underlying action. The actual capabilities depend on what the integration makes available. Model Context Protocol

Businesses should therefore look beyond whether a supplier says it “supports MCP”. A connection that retrieves a small selection of information offers a different opportunity from one that supports the complete process you want to improve.

It is also worth examining existing APIs and integration options. MCP often provides an additional layer over those capabilities, so the accessibility of the underlying application remains important. Businesses can start evaluating these foundations now without committing themselves to a particular agent platform.

Good data hygiene becomes an operational requirement

An employee can often work around poor data because they know the business. They recognise that two slightly different company names refer to the same customer, understand which spreadsheet is current and remember that a supposedly active contract ended last month.

An agent may have no dependable way to resolve those inconsistencies. If it finds two customer records with different renewal dates, it needs an authoritative source or a defined process for handling the conflict.

This makes good data hygiene particularly important when AI moves from producing answers to performing actions. An incorrect email address might lead to a message being sent to the wrong person. An outdated service record could affect a renewal proposal. An ambiguous account status could trigger an inappropriate follow-up.

Preparing data for agents therefore involves more than removing duplicates. Businesses need consistent fields, meaningful statuses, clear relationships between records and an agreed source of truth for each important type of information.

Customer identities should be linked reliably across systems. Dates and currencies should be explicit. Superseded documents should be distinguishable from current ones. Important decisions should be recorded somewhere accessible rather than remaining entirely in someone’s inbox or memory.

These improvements benefit people immediately. They also make it easier to introduce automation with confidence because there is less ambiguity for either an employee or an agent to interpret.

How businesses can get ahead now

The strongest starting point is a specific piece of work that repeatedly consumes time. Meeting preparation, internal reporting and finding information across applications are useful candidates because the results can be checked before the agent is given permission to change anything.

We would approach preparation in five stages:

Stage	Practical action
Identify the work	Choose a recurring task with a clear outcome and record how much time it currently takes.
Improve the data	Resolve duplicates, standardise key fields and agree which system holds each authoritative record.
Check connectivity	Assess relevant APIs, MCP connections, available actions and permission controls.
Define responsibility	Decide what the agent can do independently and when a person needs to review or approve it.
Test and measure	Compare results with the existing process, including errors, checking time and integration costs.

Access should follow the needs of the task. An agent preparing an account summary may need permission to read selected information, while an agent updating customer records requires a different level of control. MCP documentation describes mechanisms such as tool availability settings, approval dialogues and activity logs, but businesses still need to implement and configure suitable controls. Model Context Protocol

Testing should include the awkward cases: missing information, conflicting records, unavailable applications and requests the agent is not authorised to complete. A useful system needs to recognise those situations and make them visible.

The measure of success is the improvement to the complete process, including the time people spend checking the agent’s work. A task completed quickly has limited value if it creates a larger verification burden afterwards.

Processes need to be clear enough to delegate

Clean data alone will not explain how your business should operate.

An instruction such as “follow up overdue customers” sounds simple until someone needs to decide what counts as overdue, whether disputed invoices are excluded, who owns the relationship and which customers require a more sensitive approach.

Employees often learn these rules informally. Delegating the work to an agent requires the business to make them explicit.

That means defining the intended outcome, the information required, the actions allowed and the situations that need human judgement. It also means accepting that some parts of a process are better handled by conventional automation. A fixed calculation or validation rule can remain in software, while an agent helps gather context or manage a less predictable sequence of tasks.

For many businesses, documenting those boundaries will be one of the most useful parts of preparing for agents. It exposes assumptions and inconsistencies that may already be causing problems.

Our view

We think agents are likely to become the main way people use business software for a substantial amount of everyday work. The pace will vary, but the opportunity is clear: people can spend more time deciding, communicating and delivering, while software takes on more of the effort involved in finding information and carrying out routine actions.

Businesses can prepare without predicting which AI provider or protocol will dominate. Well-structured data, accessible applications, documented processes and sensible permissions are useful foundations across a wide range of possible futures.

The organisations that make progress on those foundations now will have more opportunities to put agents to useful work as the technology develops. They will also have better systems for their teams in the meantime.

If you want to understand where agents could help your business, Ostratto can help you review the systems and processes you already have, identify a practical starting point and put the right foundations in place.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

AI Agents are starting to act in ways their creators didn’t expect

Thu, 01 Oct 2026 06:36:00 +0000

Recent incidents involving experimental AI agents are providing an early indication of the security challenges that could emerge as artificial intelligence moves beyond generating text and begins interacting directly with websites, software and other computer systems.

One of those incidents occurred earlier this year on RubyGems, the package repository used by millions of software developers. In May, newly registered accounts began publishing large numbers of packages to the service, eventually forcing RubyGems to temporarily suspend new registrations while it dealt with the activity.

RubyGems has since confirmed that more than 500 packages were removed. Researchers investigating the campaign attributed the activity to OpenAI agents, although RubyGems says the available evidence does not allow it to independently determine whether AI agents actually created or published the packages.

The packages were designed to use shared infrastructure to execute code, retrieve publicly available information from the web and publish the results back to RubyGems. Researchers also identified attempts to obtain other users' API keys, although RubyGems says it found no evidence that those attempts were successful.

The incident attracted attention because the apparent objective wasn't simply to attack RubyGems. The agents were reportedly using the platform while attempting to complete other tasks, illustrating how an autonomous system can potentially interact with infrastructure in ways that weren't anticipated by either its developers or the people operating that infrastructure.

A much more serious incident followed

The concerns became more substantial following an incident involving OpenAI's internal AI research and Hugging Face, one of the world's largest platforms for hosting AI models and datasets.

In July, OpenAI was running cybersecurity evaluations designed to test the capabilities of experimental models. According to OpenAI's subsequent incident report, the agents were deliberately operating with fewer safeguards than its publicly available products, but they were not supposed to have unrestricted internet access or communicate freely with other agents.

Some of them nevertheless found ways around those restrictions.

Agents discovered that they could use OpenAI's Artifactory software as an unintended communication system, effectively creating a message board through which otherwise separate agents could exchange information. They also discovered ways of using infrastructure that had limited internet connectivity to make requests on their behalf.

As the evaluations continued, agents shared these techniques with others and began collaborating. OpenAI says the agents eventually exploited vulnerabilities across its own infrastructure and Hugging Face, including previously unknown vulnerabilities. During the incident, agents achieved code execution on Hugging Face servers and obtained access to credentials and some private data.

OpenAI's investigation concluded that several behaviours contributed to the incident, including reward hacking, persistence when agents encountered apparently impossible tasks, unauthorised communication and agents adopting objectives from one another. The company has described the episode as a "warning shot" and says it has subsequently introduced stricter isolation, internet restrictions, monitoring and other safeguards.

AI is beginning to operate software rather than simply discuss it

These incidents took place in unusual research environments and involved experimental models operating under conditions that differ significantly from the AI products most businesses use today. They shouldn't therefore be interpreted as evidence that an ordinary AI assistant connected to a CRM is likely to behave in the same way.

They are significant, however, because the technology industry is rapidly moving towards giving AI systems more autonomy.

Most current business use of generative AI still involves a person asking a question and receiving an answer. Even when AI is used to write an email, analyse a spreadsheet or summarise a customer record, a person will usually decide what happens next.

AI agents are intended to take this further by allowing a model to use tools and software to complete a task. An agent might browse websites, query databases, call APIs, update CRM records, work with documents or communicate through other applications as part of a longer process.

For businesses, this could make AI considerably more useful. An agent connected to a CRM, for example, could potentially identify customers requiring attention, review their history, prepare correspondence, create follow-up tasks and update records without somebody manually moving between applications.

Connecting the same agent to email, accounting software, cloud storage and other systems expands what it can achieve, but it also increases the consequences when the system makes an incorrect decision.

The risk is not limited to deliberately malicious AI

The incidents also illustrate why discussions about AI safety don't necessarily need to involve a system deliberately attempting to cause harm.

A sufficiently autonomous system could cause problems simply because it interprets an objective differently from its developers, discovers an unexpected way of completing a task or interacts with another system in a way that wasn't considered when either system was designed.

This is particularly relevant to business automation. Traditional automation is usually based on relatively predictable rules. A workflow might be configured so that when a particular event occurs in a CRM, a defined action follows.

An AI agent can operate differently because it may be given an objective and allowed to determine the individual steps required to achieve it. That flexibility is precisely what makes agents attractive, particularly for processes that are too complicated or variable for conventional automation, but it also makes their behaviour less predictable.

In a normal business environment, an unexpected action could mean incorrectly modifying CRM records, communicating with the wrong customers, accessing information that isn't relevant to the task, creating transactions or triggering other automated processes.

None of those scenarios requires an AI system to be intentionally hostile. They simply require the system to have sufficient access to turn a poor decision into an action.

Businesses will need to think carefully about AI permissions

Cybersecurity already has a well-established concept for dealing with this problem. The principle of least privilege means that a person or application should only receive the permissions genuinely required to perform its role.

The same approach is likely to become increasingly important for AI agents.

An agent analysing CRM records may need permission to read customer information without requiring the ability to delete it. An agent helping with invoicing might be allowed to prepare transactions while requiring human approval before they are sent. An AI researching suppliers might require access to the internet without needing unrestricted access to internal documents.

Businesses will also need to decide which actions remain subject to human approval. Financial transactions, deleting information, changing permissions, publishing content and communicating externally are obvious areas where allowing an agent to act completely independently deserves careful consideration.

Monitoring will be equally important. If an AI is able to act on behalf of an organisation, businesses need reliable records of which systems it accessed, what it changed and which actions it performed. OpenAI's response to the Hugging Face incident includes greater monitoring of agent behaviour alongside stricter isolation and access controls, reflecting how important those safeguards become as models gain greater autonomy.

A new phase of AI adoption

AI agents could ultimately prove much more significant to businesses than the chatbots that introduced generative AI to most people. Allowing software to understand a broader objective and work across several applications could automate processes that have traditionally required considerable human administration.

The events at RubyGems and Hugging Face don't undermine that opportunity, but they do provide useful evidence of why access and permissions need to develop alongside the technology.

The security discussion around AI has so far concentrated heavily on the information models produce, including inaccurate answers, confidential information and generated content. Agentic AI adds another consideration because the model may increasingly be able to act on the information it produces.

As businesses begin connecting AI to CRMs, finance platforms, email accounts, cloud storage and other operational systems, deciding what an agent is permitted to do may become just as important as deciding which AI model to use.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

How much water does AI actually use?

Mon, 21 Sep 2026 08:20:00 +0000

Every time someone asks ChatGPT a question, generates an image or uses AI to analyse a document, computers inside a data centre process the request.

Those computers generate heat. Keeping them cool can require water, while generating the electricity that powers them may use even more.

For one simple AI prompt, the amount can be tiny. Across billions of prompts and increasingly powerful data centres, the figures become much harder to ignore.

The short answer

Google estim ates that a median Gemini text prompt consumes approximately 0.26 millilitres of water, or around five drops.

At that rate, you could submit nearly 2,000 prompts before using the equivalent of a 500 ml bottle of water. Even 10,000 prompts would account for approximately 2.6 litres.

However, earlier academic research estimated that GPT-style models could consume around 500 ml for every 10 to 50 responses.

The estimates differ because they examine different models, locations, cooling systems and sources of water use. AI infrastructure has also become considerably more efficient.

The larger figures reveal the real scale:

Global data centres consume more than 560 billion litres annually
A large 100 MW data centre could consume around 2.5 billion litres per year
Global data centre consumption could reach 1.2 trillion litres annually by 2030
AI-related water withdrawal has been projected to reach 4.2 to 6.6 trillion litres in 2027

Not all data centre activity relates to AI, and water withdrawal is different from water permanently consumed. Nevertheless, AI is helping to drive the rapid expansion of this infrastructure.

Why does AI use water?

AI’s water footprint comes from three main sources.

The first is data centre cooling. The processors used to run AI models generate substantial heat. Some data centres use evaporative cooling, where water absorbs the heat before evaporating. A UK report on water use in AI and data centres suggests that around 80% of the water used by these systems can be lost through evaporation.

The second is electricity generation. Many power stations use water for cooling and other processes, creating an indirect water footprint outside the data centre.

The third is chip manufacturing. Semiconductor factories use highly purified water to clean silicon wafers while producing the advanced processors on which AI depends.

Some data centres use air cooling, closed-loop systems or reclaimed wastewater, so the amount of drinking-quality water required varies significantly between facilities.

Data-center Cooling

Water removes heat generated by servers. In evaporative cooling systems, much of that water is lost to the atmosphere.

Electricity Generation

Generating the power used by AI can consume water away from the data centre, creating an indirect water footprint.

Chip Manufacturing

AI servers rely on advanced processors, which require ultrapure water to clean silicon wafers during manufacturing.

Not every AI request is equal

A short text question does not require the same computing power as generating a video, analysing hundreds of pages or asking an advanced model to work through a complex problem.

Water use can be affected by:

The model selected
The length of the input and response
Whether extended reasoning is required
The number and size of uploaded files
Whether the system generates text, images, audio or video
The efficiency of the processors
The cooling system and local weather
The source of electricity

The International Energy Agency reports that the energy needed for individual AI tasks has been falling rapidly. However, video generation, extended reasoning and autonomous agents can consume hundreds or thousands of times more energy than simple text generation.

AI is becoming more efficient, but it is also being used more frequently and for more demanding work.

What does a large data centre use?

A 100 MW hyperscale data centre could consume approximately 2.5 billion litres of water annually, depending on its design and location.

That is equivalent to:

1,000 Olympic swimming pools every year
Nearly 6.9 million litres every day
The average daily household use of around 50,000 people in England

Not every data centre uses this much. Facilities using air cooling or recycled water may use considerably less. Climate also makes a difference, with hotter locations generally requiring more cooling.

The figure still demonstrates the difference between one prompt and the infrastructure processing millions of them around the clock.

From 224,000 to 480,000 swimming pools

Data centres globally currently consume more than 560 billion litres of water each year.

That would fill approximately:

224,000 Olympic swimming pools per year
More than 600 Olympic swimming pools every day
The equivalent annual domestic water use of over 11 million people in England

Not all of this is caused by AI. Data centres also support websites, cloud storage, streaming services and business software.

However, AI is contributing to the construction of new data centres and the installation of more powerful computing equipment. By 2030, global data centre water consumption could reach 1.2 trillion litres annually, equivalent to approximately 480,000 Olympic swimming pools.

That is more than 1,300 swimming pools every day, or almost one every minute.

How much do technology companies use?

In 2022, Google’s data centres consumed approximately 19.5 billion litres of water, enough to fill around 7,800 Olympic swimming pools.

Microsoft reported global water consumption of approximately 6.4 billion litres during the same year, representing an increase of 34% from 2021. That equates to roughly 2,560 swimming pools.

These totals cover wider cloud and corporate operations rather than AI alone. They do, however, show why per-prompt efficiency only tells part of the story.

A company can reduce the water required for each AI response while increasing its total consumption because the number of users, services and data centres is growing faster.

Could AI withdraw trillions of litres?

Academic research has projected that global AI demand could be responsible for withdrawing between 4.2 and 6.6 trillion litres of water in 2027.

At the upper end, that is enough to fill approximately 2.6 million Olympic swimming pools.

This figure relates to water withdrawal, which is the amount taken from a water source. Some of that water may later be treated and returned. Water consumption refers to the portion that evaporates or is otherwise unavailable for immediate reuse.

The figure therefore does not mean that 6.6 trillion litres will disappear. However, withdrawing water can still affect local supplies, particularly if it is returned somewhere else, at a different temperature or after the period when it was most needed.

Why this matters in the UK

The average person in England uses 136.5 litres of water per day. For comparison, a bath uses around 80 litres, a standard ten-minute shower can use between 100 and 150 litres, and a washing-machine cycle typically uses around 50 litres.

England is forecast to face a daily water deficit of nearly five billion litres by 2050. That is equivalent to the current daily household use of more than 36 million people.

Several existing and proposed data centre clusters are in areas already considered seriously water stressed. Cooling demand may also be highest during hot weather, precisely when household, agricultural and environmental demand is under the greatest pressure.

Water use is therefore not just about the global total. Where and when it is consumed matters enormously.

Should businesses use AI less?

The answer is not necessarily to stop using AI. A useful prompt that saves several hours of work may be a reasonable use of resources.

The better question is whether AI is being used deliberately or added to processes without a clear benefit.

Businesses can make more responsible choices by selecting smaller models for straightforward tasks, limiting unnecessary image and video generation, avoiding repeated requests without a purpose and considering environmental transparency when choosing technology providers.

Responsibility should not sit entirely with individual users. Technology companies need to report water use more consistently, including how much drinking-quality water they consume, where it is used and which cooling technologies they deploy.

Water use is falling

AI models, processors and data centres are becoming more efficient. A simple prompt can now require less energy and cooling water than it once did, reducing the environmental impact of each individual task.

AI usage is rising

At the same time, more people and businesses are adopting AI for increasingly complex work. The growth in users, prompts, image generation and other demanding tasks can outweigh the savings achieved per interaction.

Five drops can still become a flood

It is misleading to suggest that every short AI conversation consumes an entire bottle of water. Recent provider data indicates that a straightforward text prompt can use only a fraction of a millilitre.

It would be equally misleading to use that figure to dismiss the industry’s wider impact.

Data centres already consume hundreds of billions of litres each year. That demand could reach 1.2 trillion litres by 2030 as AI use grows and applications become more demanding.

One useful prompt is unlikely to create a meaningful water problem. Billions of prompts and enormous data centres concentrated in water-stressed areas potentially can.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

Could AI actually be a risk to humanity?

Sun, 13 Sep 2026 20:48:00 +0000

There is something unusual happening in artificial intelligence. Some of the people working on the world's most advanced AI systems are warning that the technology they are helping to develop could eventually pose a catastrophic risk to humanity. Others within the same field argue that these fears are exaggerated, based on speculative assumptions about technologies that do not yet exist.

That disagreement might be easier to dismiss if AI development were progressing slowly. It isn't. The largest technology companies are spending extraordinary amounts of money on computing infrastructure, new models are arriving regularly, and AI systems are being given greater autonomy and access to tools that allow them to do more than simply generate text.

Governments have responded by establishing AI safety institutes and developing new regulations, while researchers are testing whether increasingly capable models can operate independently, deceive evaluators or perform some of the tasks that would theoretically be required to replicate themselves.

The debate has now reached some unlikely places. King Charles is hosting an AI summit at Dumfries House, bringing together senior figures from the technology industry to discuss the future development of artificial intelligence and its impact on society.

Against that background, the question of whether AI could eventually present a serious risk to humanity no longer feels quite as easy to dismiss as science fiction. The difficult part is working out how seriously we should take a danger that remains highly uncertain.

How worried are the people building it?

In September 2026, concerns about AI risk received renewed attention following warnings from researchers working close to the technology. In an interview examining whether AI could realistically threaten humanity within the next decade, The Conversation discussed an estimate from Anthropic alignment researcher Evan Hubinger that there was a greater than 10% probability of AI killing humanity within that period.

That figure needs considerable context. Humanity has never built a superintelligent AI, so there is no historical evidence from which anyone can reliably calculate such a probability. Professor Kate Devlin of King's College London points out that predictions of human-level or superhuman AI being only a few years away have been made before, and nobody actually knows whether today's technology will ultimately lead there.

But uncertainty doesn't necessarily make the issue irrelevant. When the possible consequences are sufficiently severe, even a small and uncertain probability can justify precautions. The problem with AI is that we're being asked to make those judgements while the technology itself is changing extremely quickly.

What could actually go wrong?

The popular version of an AI catastrophe usually involves machines becoming conscious and turning against humans. Most serious AI safety research is considerably less cinematic.

One of the more immediate concerns is misuse. The 2026 International AI Safety Report examines the potential for increasingly capable AI to assist with cyber attacks and provide information relevant to biological and chemical weapons. There are still substantial practical barriers between an AI providing information and somebody successfully causing serious harm, but more capable systems could gradually make specialist knowledge and complex processes easier to access.

AI also doesn't necessarily need to be deliberately misused. As we move from chatbots that provide answers towards AI agents that can access browsers, files, software and other systems, mistakes become more consequential. An incorrect chatbot response can be ignored; an autonomous system acting on an incorrect assumption may be able to make changes before a human intervenes.

The most extreme scenario is a future AI sufficiently capable and autonomous that humans struggle to control it. Today's systems aren't capable of this, something the International AI Safety Report makes clear. What interests researchers is whether some of the capabilities that might eventually contribute to such a scenario are beginning to emerge.

Some of those capabilities are improving quickly

The UK's AI Security Institute tests frontier AI models against tasks associated with autonomy, cyber security and other potentially dangerous capabilities.

One particularly striking example involves simplified self-replication evaluations. These are controlled tests of individual skills that might be required for an AI to obtain computing resources and maintain access to them. In 2023, the strongest models achieved success rates below 5% across many of these evaluations. By 2025, frontier models had achieved more than 60%.

That does not mean an AI currently has a 60% chance of escaping onto the internet and copying itself. AISI says current models still struggle with important parts of the process and there is no evidence that models are spontaneously attempting to self-replicate.

The significance is the speed at which capabilities are changing. AISI has found rapid improvements across the areas it measures, with performance in some categories doubling approximately every eight months.

That doesn't demonstrate that an AI catastrophe is approaching. It does explain why some researchers are uncomfortable waiting for definitive evidence of danger before taking precautions.

There are good reasons to be sceptical

Existential AI risk shouldn't be treated as established fact simply because prominent people within the industry are discussing it.

As Professor Kate Devlin argues, we don't currently have superintelligence and nobody knows when, or even whether, it will arrive. Today's systems remain unreliable in surprisingly basic ways, and there is no guarantee that making current models larger will eventually create something vastly more intelligent than humans.

There are commercial incentives worth considering too. AI companies benefit from the perception that the technology they're developing is extraordinarily powerful. Describing AI as potentially civilisation-changing can attract investment, political attention and customers as effectively as it can generate genuine concern.

There is also a risk that hypothetical extinction scenarios distract attention from problems already happening. AI-assisted fraud, misinformation, copyright disputes, environmental impacts and disruption to employment don't require superintelligence.

It shouldn't have to be one debate or the other. We can address the measurable harms of today's AI while investigating the potentially much larger risks of future systems.

The bigger risk may be the race

Perhaps the strongest reason for taking AI safety seriously is the incentive structure surrounding its development.

OpenAI, Anthropic, Google, Meta and others are competing for customers, investment, computing power and technical talent. Governments increasingly regard AI leadership as strategically important too.

That makes slowing down difficult. If one company delays a new system because it believes further safety research is required, a competitor may continue. The same problem exists between countries. Cautious development might benefit everyone collectively while rapid development remains the rational choice for each individual participant.

This is behind some of the calls for international intervention. Writing in The Guardian, Gaby Hinsliff argues for pausing particularly risky areas of AI research while international safeguards are developed, rather than stopping AI research altogether.

That distinction matters. We already accept that potentially transformative technologies can require significant restrictions without abandoning them. Pharmaceuticals undergo clinical trials, aircraft require certification and nuclear technology is subject to extensive controls. AI is different from all of those technologies, but the principle that greater potential consequences justify greater scrutiny isn't particularly radical.

So, could AI actually threaten humanity?

There is currently no good reason to believe that ChatGPT, Claude, Gemini or another existing AI system is about to destroy humanity. Current models lack the autonomy and broader capabilities required by the more extreme scenarios being discussed.

There is equally little basis for confidently saying that sufficiently advanced AI could never become dangerous. We don't know how capable these systems will become, how quickly those capabilities will emerge or whether the safety techniques being developed alongside them will prove sufficient.

The 2026 International AI Safety Report reflects this uncertainty rather well. Its contributors don't agree on the probability of humanity losing control of future AI systems and the report doesn't predict that such an outcome will happen. It does, however, recognise that the possible consequences are serious enough to justify research and preparation.

The fact that King Charles is now bringing some of the industry's biggest names together to discuss the future of AI is another indication of how far this discussion has moved beyond technology companies and research laboratories.

Perhaps today's fears about superintelligence will eventually look misplaced. Current AI architectures might reach their limits and the predicted leap towards vastly more capable systems may never happen.

But we're currently developing the technology while simultaneously trying to understand what its limits and risks might be. The important question isn't whether anyone can prove that AI will threaten humanity, because nobody can. It's how much evidence of that possibility we should require before deciding that some precautions are worthwhile.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

What is MCP, and what does it mean for businesses?

Tue, 01 Sep 2026 08:00:00 +0000

Artificial intelligence has become considerably more capable over the past few years, but there is still a fairly fundamental limitation to the way most businesses use it. An AI assistant might be able to analyse a document, write an email or answer a complicated question, but it usually has very little awareness of what is actually happening across the organisation using it.

The useful information is somewhere else. Customer relationships and sales activity might be held in a CRM, financial information in an accounting platform, ongoing work in a project management system and company knowledge across documents, emails and other applications. Unless that information is deliberately provided to the AI, it has little or no context from which to work.

Model Context Protocol, or MCP, is one of the technologies attempting to solve that problem. It provides a standard way for AI applications to connect to external systems and use the information and capabilities available within them. While the name makes it sound like something primarily of interest to developers, its wider implications are much more relevant to businesses.

MCP could help change AI from a tool that sits alongside business software into something capable of actually working across it.

What is MCP?

MCP is an open standard for connecting AI applications to external data sources and tools. It was originally introduced by Anthropic in 2024 and has subsequently developed into a broader industry standard, with support emerging across many of the companies building AI models, development tools and business applications.

One of the easiest comparisons is USB. Computers can interact with thousands of different devices because manufacturers have agreed on common standards for connecting them. A keyboard manufacturer does not need to design a completely different connection for every make of computer, and computer manufacturers do not need to develop proprietary connections for every possible peripheral.

MCP attempts to introduce a similar principle to AI integrations.

An application can make information and functionality available through an MCP server, while an AI application capable of using MCP can connect to that server and discover what is available. Rather than developing an entirely bespoke connection between every AI product and every business application, MCP provides a common language through which those connections can be made.

It is important not to confuse this with MCP replacing APIs. In many cases, an MCP server will itself use an application's existing API. MCP instead provides a standardised layer that makes those capabilities understandable and usable by AI systems.

How does MCP actually work?

At a basic level, an MCP connection involves an AI application acting as a client and an MCP server exposing information or capabilities that the AI can use. The server effectively tells the AI what it has access to and how those capabilities can be used.

Imagine, for example, that an organisation connects its CRM to an AI assistant through MCP. Rather than exporting sales data into a spreadsheet and uploading it to ChatGPT or another AI tool, someone could simply ask which opportunities have not been contacted during the past fortnight. The AI could retrieve the appropriate records from the CRM and produce an answer based on live information.

The more significant change comes when the AI is allowed to perform actions as well as retrieve information. Having identified those neglected opportunities, the user might ask the AI to create follow-up tasks for each account owner. Provided the relevant tools and permissions have been made available, the AI could carry out the request in the CRM rather than merely explaining how the user should do it.

This distinction between retrieving information and performing actions is central to understanding why MCP is attracting so much attention. Connecting AI to business systems is useful because it gives the model better context, but connecting it to the functions within those systems opens up the possibility of AI becoming an active participant in business processes.

From answering questions to completing work

Most business use of generative AI still revolves around producing something for a person. An employee asks for an email, summary, report, calculation or piece of analysis, receives an answer and then decides what to do with it.

That model is already useful, but it leaves the employee responsible for moving between the AI and the applications where the work actually happens. If an AI assistant identifies that ten sales opportunities require attention, somebody still has to open the CRM, find those records and create the necessary tasks.

MCP is part of a broader move towards AI agents, where the objective is not simply to generate an answer but to allow AI systems to use tools in order to complete a task. Once an AI can securely access the systems a business already uses, the range of useful tasks becomes considerably wider.

Consider a fairly ordinary request such as asking an AI assistant for an update on a particular customer. A useful answer might require information from several places. The CRM contains the customer's sales history and current opportunities, the finance system contains invoices and payment information, the support platform contains outstanding issues and the project management system shows the progress of ongoing work.

Today, an account manager might open each of those applications and piece together the information manually. With appropriate MCP connections, an AI agent could potentially retrieve the relevant information from each system and produce a single account summary.

The employee could then ask it to create a task for an overdue sales follow-up, prepare an email regarding an outstanding invoice or summarise an unresolved support issue. Rather than AI being another application that needs to be checked, it starts to become a way of interacting with the applications the business already has.

But haven't we been connecting applications for years?

None of this means that integrations are new. Businesses have been connecting applications through APIs, webhooks and integration platforms for a long time, and those technologies will continue to be important.

The difference is largely in how the integration is used.

A traditional automation is normally designed around a predefined process. When an opportunity reaches a particular stage in the CRM, create a project. When an invoice is paid, update another record. When a form is submitted, send the information to another application. The trigger, action and expected outcome are defined in advance.

An AI agent can be much less rigid. A person can describe an objective in normal language, after which the AI can determine which available tools it needs in order to complete the request. Depending on the task, that might involve retrieving information from one system, comparing it with information from another and then performing an action somewhere else.

MCP helps make this possible by giving the AI a consistent way to discover what tools and information are available. It does not eliminate the underlying integrations, but it can make them accessible to AI in a much more flexible way.

Why does having a standard matter?

Without a common standard, connecting AI applications to business software quickly becomes complicated. An integration built between one AI assistant and a CRM might be completely different from the integration required by another AI assistant. Multiply that across accounting, support, project management, document storage and internal systems, and organisations end up maintaining a growing collection of proprietary connections.

A widely adopted protocol changes that model. Software providers and developers can expose capabilities through MCP, while AI applications can implement support for the same protocol. It does not mean that every AI application automatically gains access to every business system, nor does it remove the need for authentication and configuration, but it provides a common foundation on which those connections can be built.

This is arguably one of the reasons MCP matters more than many of the individual AI features being released at the moment. Features come and go relatively quickly, whereas standards can become part of the underlying infrastructure that allows an ecosystem to develop.

What could MCP actually mean for a business?

The immediate opportunity is better access to information. AI is considerably more useful when it can work with accurate organisational context rather than relying entirely on information pasted into a conversation.

A sales manager could ask questions about the live pipeline without first preparing a report. A finance team could investigate transactions without exporting data into spreadsheets. A project manager could ask which projects are at risk based on deadlines, tasks and recent activity. A support manager could identify recurring customer issues by allowing an AI system to analyse ticket information directly.

However, the larger opportunity is the ability to act on what the AI discovers. Depending on the systems connected and the permissions granted, an agent could create or update CRM records, generate tasks, prepare documents, organise information, update projects or initiate established business processes.

That does not necessarily mean handing entire departments over to autonomous agents. In many businesses, the most useful applications are likely to be much less dramatic. Removing ten minutes of repetitive administration from a process that happens hundreds of times a month can have considerably more practical value than building an impressive AI demonstration that nobody trusts enough to use.

AI could change how we interact with business software

There is also a longer-term implication that is easy to overlook. For decades, software has largely been designed around graphical interfaces. If somebody wants to update an opportunity, they log into the CRM, navigate to the appropriate record, find the correct field, make the change and save it.

An AI agent with access to the same system introduces another interface. Instead of navigating through the application, someone might ask it to move an opportunity to the negotiation stage, add a note explaining that legal approval is outstanding and create a reminder to follow up next Tuesday.

The CRM has not disappeared. It still stores the information, controls the business logic and provides the underlying functionality. What has changed is the way the employee interacts with it.

If that becomes commonplace, the characteristics businesses value in software may also begin to change. A polished user interface will remain important, but so will the accessibility and structure of the data underneath it. Applications with comprehensive APIs, well-designed permissions and good interoperability become particularly valuable when people are no longer the only users interacting with them directly.

In that environment, the quality of a company's underlying systems may matter more than the particular AI model it chooses to put in front of them.

Security and permissions become much more important

There is an obvious consequence to giving AI systems access to business applications. An AI that can retrieve information from a CRM represents one level of risk; an AI that can modify or delete information represents another.

MCP does not remove the need for conventional security controls. Organisations still need to consider authentication, permissions, data access, audit logs and which actions should require approval. The fact that an AI agent is performing an action rather than an employee does not make those controls any less important.

In practice, businesses are likely to adopt these capabilities gradually. An AI system might initially receive read-only access to particular information, followed by permission to perform a limited set of low-risk actions. More consequential changes could continue to require explicit human approval.

The objective should not be to give an AI access to everything simply because it is technically possible. It should be to give it the minimum information and functionality required to perform a useful task safely.

AI won't fix poor systems

MCP also highlights something that is sometimes lost in discussions about AI transformation: connecting an intelligent model to poor-quality business data does not suddenly produce a good business system.

If a CRM contains duplicate records, inconsistent information and badly designed processes, an AI agent will have to work with those same problems. If nobody knows which system contains the authoritative version of a customer's information, providing an AI with access to all of them may create more ambiguity rather than less.

Businesses that have invested in structured data, documented processes, sensible permissions and properly connected systems are therefore likely to be in a much better position to benefit from AI agents.

This is one reason we think the discussion around AI needs to extend beyond simply choosing an AI product. The underlying systems still matter. In fact, as AI becomes capable of doing more with them, they may matter even more.

What should businesses be doing about MCP now?

For most businesses, the answer is not to start building MCP servers for every application tomorrow. The technology and the wider ecosystem are still developing quickly, and there is little value in adopting a protocol simply for the sake of being early.

There is, however, a good reason to start considering how ready your organisation is for this type of technology.

When evaluating software, businesses should increasingly consider how accessible their data is, whether the application has a comprehensive API, how granular its permissions are and how easily it can interact with other systems. Internally, it is worth identifying processes that involve employees repeatedly finding information, copying it between applications or carrying out predictable administrative actions.

Those processes are likely to be some of the strongest candidates for useful AI automation as the technology matures.

More importantly, businesses should continue improving the quality and structure of their underlying systems. Whether MCP ultimately becomes the dominant standard or is eventually replaced by something else, AI systems are clearly moving towards greater access to external tools and organisational data. Businesses with well-organised systems will be in a much stronger position to take advantage of that change.

The important part isn't really MCP

MCP is ultimately a technical standard, and most employees will probably never need to know whether they are using it. In much the same way that somebody does not need to understand the protocols behind the web to use a website, the technical details should eventually disappear behind the applications people use every day.

What matters is the change it represents.

Until now, much of the value of generative AI has come from what a model already knows and whatever information a user provides during a conversation. The next stage is increasingly about giving AI controlled access to the information and tools that already exist within organisations.

If that develops in the way many technology companies expect, the role of AI within businesses could change substantially. Rather than being another piece of software employees open when they need help writing or analysing something, AI could become a layer through which employees interact with much of their existing technology.

At that point, the most important question for a business will no longer be simply which AI model it wants to use. It will be whether its systems, data and processes are in good enough shape for AI to work with them.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

The AI Bubble Will Burst. That Might Be a Good Thing.

Sat, 22 Aug 2026 11:04:04 +0000

AI is already useful. It writes and reviews code, summarises documents, helps answer customer enquiries, analyses data and is steadily appearing inside the software businesses already use every day. We use it ourselves, and we think it is going to have a significant impact on the way businesses operate.

But believing AI is useful is not the same thing as believing that every investment being made around AI makes financial sense.

There is an enormous amount of money flowing into AI companies, chips, data centres and the energy infrastructure needed to power them. Stanford's latest AI Index found that global corporate AI investment more than doubled in 2025, while investment in generative AI grew even faster. At the same time, the biggest technology companies are committing extraordinary sums to the infrastructure they believe will be needed to satisfy future demand.

That demand may well arrive. In fact, there is plenty of evidence that some of it already has. The problem is that today's investment does not merely assume that AI will be popular. It assumes it will become valuable enough, quickly enough, to justify hundreds of billions of dollars being spent ahead of it.

That is where the bubble starts to appear.

What do we actually mean by an AI bubble?

Calling something a bubble can make it sound as though the underlying thing is fake. That is not really what a financial bubble means.

Strictly speaking, economists generally use the term for a situation where the price of an asset moves beyond what its underlying fundamentals can reasonably justify. Bubbles can become self-reinforcing: rising prices attract investment, that investment creates further growth and confidence, and the apparent success encourages even more money to enter. Eventually the fundamentals have to catch up with the expectations. If they don't, the process can reverse.

With AI, we are using the word slightly more broadly. The concern is not simply that a handful of AI shares are expensive. It is that valuations, infrastructure spending and business plans across an entire industry increasingly depend on very optimistic assumptions about how much AI customers will eventually consume and how much they will be prepared to pay for it.

That distinction matters because a technology can be genuinely transformative and still have a financial bubble built around it.

The dot-com boom is the obvious example. Investors were right that the internet would change the world. Where they went wrong was assuming that almost every internet company, at almost any valuation, would become a great business. When that optimism collided with reality, the Nasdaq fell around 77% from its 2000 peak to October 2002 and many internet companies disappeared. The technology itself continued to develop.

AI could follow a similar pattern.

Where the AI bubble is actually forming

The easiest way to see the scale of it is to follow the money into infrastructure.

Microsoft said in July that its calendar-year 2026 capital expenditure expectation was around $175 billion after an accounting-related lease change. It spent $41 billion in its latest quarter alone, with roughly two-thirds going towards relatively short-lived assets such as CPUs and GPUs. Amazon has said it expects to invest about $200 billion across the business in 2026, while its latest results said the large increase in property and equipment purchases primarily reflected AI investment. Meta expects capital expenditure of between $130 billion and $145 billion this year.

These aren't obscure AI startups gambling venture capital. They are some of the largest and most profitable companies in the world, building physical infrastructure at an extraordinary pace.

And to be fair, there is real demand behind it. AWS revenue grew 37% year-on-year in Amazon's latest quarter, Microsoft said Azure revenue grew 43%, and Microsoft 365 Copilot has passed 30 million paid seats. This is one reason we don't buy the argument that AI is simply a fad waiting to disappear.

The more interesting question is whether the eventual demand is large enough to justify the amount being built for it.

There is also an unusual dynamic inside the AI economy. The companies building the infrastructure are not always simply selling computing power to independent customers. They are also investors and strategic partners in some of the AI companies consuming huge quantities of that infrastructure. Microsoft's financial results, for example, now separately discuss the effects of its OpenAI investment, while Amazon's latest quarter included a very large accounting gain from its investments in Anthropic.

That doesn't make the revenue imaginary, nor does it prove that the whole market is circular. But it does make the industry more complicated than the simple story of millions of ordinary businesses suddenly buying hundreds of billions of dollars' worth of AI.

The argument in Ed Zitron's recent piece about an "AI demand bubble" takes this much further, arguing that the industry is dangerously dependent on a relatively small number of heavily funded AI labs buying enormous amounts of computing capacity. We wouldn't treat the analyst estimates used to make that case as established fact, particularly because the cloud companies do not disclose enough detail to verify them. But the broader question it raises is a sensible one: where does the money eventually come from once you follow the chain all the way back to the end customer?

OpenAI and Anthropic are projected to account for more than 70% of Amazon’s AI revenues. Source: Barclays research, via Where’s Your Ed At.

The awkward question is what customers will actually pay for

Business adoption of AI is unquestionably growing. The more difficult part is proving what that adoption is worth.

A recent NBER study surveyed nearly 6,000 senior executives across the UK, US, Germany and Australia. It found that 69% of firms were already using AI in some form. But 89% of executives said AI had made no measurable difference to labour productivity at their business over the previous three years. Among the minority reporting an impact, the researchers estimated the average productivity increase across all firms at just 0.29%. Executives expect substantially larger benefits in the years ahead, which may turn out to be correct, but that gap between what AI has delivered so far and what businesses expect it to deliver next is exactly where the financial risk sits.

That feels reasonably consistent with what we see in the technology industry. There are AI products that are genuinely excellent and already difficult to imagine giving up. There are also a lot of software companies that appear to have added an AI assistant because somebody decided the product needed an AI button.

Eventually, customers become less interested in whether something contains AI and start asking whether it saves enough time, makes enough money or improves something enough to justify another subscription.

There is another complication. Generative AI has a real usage cost. Model providers charge according to tokens, calls, generated media and other forms of consumption because every interaction ultimately requires computing resources. That is different from software where adding another active user can sometimes cost very little. AI can certainly become cheaper to run, and it already has in many cases, but falling inference costs are a double-edged sword: they make AI easier to deploy while also making it harder to justify premium prices for products that are little more than a thin layer over somebody else's model.

This means the industry has two things to prove at once. AI needs to become valuable enough that customers will pay materially more for it, while providers need to make it efficient enough that serving those customers produces attractive margins.

That is achievable for some products. It is much harder to believe it will be achievable for all of them.

What would the bubble bursting actually look like?

We don't think there will necessarily be a morning where everybody collectively decides AI was a mistake.

A more believable version is much less dramatic. Growth remains strong, but not quite as strong as investors expected. Businesses become more selective about the AI tools they pay for. A few heavily funded AI companies struggle to justify another enormous funding round. Some data-centre projects are delayed. Investors start demanding profits rather than promises of future scale.

Once expectations change, the same feedback loop that helped inflate the market can begin to work in reverse.

Private AI businesses that could previously raise money based on what they might become in five years suddenly have to explain how they make money today. Dozens of companies selling broadly similar AI products consolidate or disappear. Features that once commanded a £20 or £30 monthly add-on get bundled into existing software. Model providers compete harder on price, pushing the underlying technology towards becoming a commodity.

The infrastructure companies have more room to adjust than is sometimes implied. Microsoft explicitly told investors in July that if demand changes it can slow purchases of shorter-lived CPUs and GPUs and stagger data-centre build-outs. That flexibility reduces the chance of a single catastrophic collapse, but the fact that investors are already asking Microsoft how it would respond to AI overcapacity shows that the possibility is no longer a particularly fringe concern.

So a burst may look less like AI suddenly disappearing and more like a fairly brutal repricing of who actually captures value from it.

There would still be GPUs running, data centres operating and businesses using AI. There would simply be fewer companies able to justify enormous valuations on the assumption that they will eventually own a large part of the market.

The dot-com comparison is useful, but not for the usual reason

The lesson of the dot-com bubble isn't "new technology gets overhyped and fails". It is almost the opposite.

The internet really did transform commerce, communication, entertainment and almost everything else. But being correct about the importance of the internet did not automatically make investors correct about which businesses would win, how quickly the opportunity would develop or what those businesses were worth in 1999.

A huge amount of capital was destroyed when those assumptions were reset. At the same time, the infrastructure, knowledge and consumer habits developed during the boom did not vanish. Companies that built genuinely useful businesses on top of the internet went on to become far larger afterwards.

That is the comparison we think makes sense for AI. It is completely possible that today's AI boom is simultaneously producing one of the most important technologies of our generation and a spectacular amount of bad investment.

In fact, those two things often go together. A genuinely important technological shift makes an especially convincing investment story because there is something real underneath the hype. Nobody has to invent the potential of AI. You only have to exaggerate how quickly that potential turns into revenue, or assume that your particular company will be one of the businesses collecting it.

AI after the bubble could be much more useful

If there is a correction, we suspect the AI industry that comes out of it will look considerably less exciting from the outside.

There will probably be fewer standalone AI products doing essentially the same thing. The underlying models will become cheaper and more interchangeable. More AI functionality will simply be included in the software businesses already use, rather than sold as an entirely separate category.

Most importantly, companies will gradually stop caring about whether they are "using AI". They will care that a process that used to take three hours now takes twenty minutes. They'll care that staff can find information more quickly, that a salesperson spends less time updating the CRM, that customer enquiries are dealt with faster or that somebody no longer has to copy information between three different systems.

That is when AI starts becoming infrastructure rather than a marketing category.

It is also why we don't think businesses should respond to talk of an AI bubble by avoiding AI. The sensible response is almost the opposite. Experiment with it, use it where it solves a genuine problem and measure whether it actually improves something. Just don't assume that every company with "AI" in its pitch deck will still exist in five years, or that every new AI subscription deserves a permanent place in your software stack.

There is plenty of evidence that demand for AI is real. There is also evidence that the measurable economic benefits are still catching up with the extraordinary level of investment being made in anticipation of them.

Our view is that those two things eventually have to meet somewhere in the middle. Expectations will come down, weaker businesses will disappear, infrastructure spending will become more disciplined and the companies left standing will have to prove that their products create enough value for customers to keep paying for them.

That would be painful for plenty of investors and AI companies. For everyone who actually wants to use the technology, though, it might be one of the healthiest things that could happen.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

AI in 2026: Five Practical Predictions for Small Businesses

Fri, 14 Aug 2026 08:44:32 +0000

AI predictions have a habit of going one of two ways. Either they promise that everything is about to change overnight, or they are so vague that almost anything can count as being right.

The reality in 2026 is a little more useful.

A recent Wise article highlighted five major themes for the year: AI agents, businesses creating more tailored AI systems, measurable returns from specific types of work, AI becoming embedded into existing software, and the continued importance of human oversight. Six months into the year, there is already plenty of evidence to test those ideas against.

For small businesses, the important question is not whether every prediction comes true. It is where AI can genuinely save time, improve a process or remove repetitive work without creating a bigger problem somewhere else.

Here is our view.

1. AI agents will start doing jobs, not just answering questions

Most people first experienced generative AI as a chatbot: ask a question, get an answer.

Agents go a step further. Instead of simply telling you what to do, an agent can potentially complete a sequence of steps across different systems. That could mean reading a support request, checking the customer record, looking up an order, applying a company policy and preparing or carrying out an approved action.

That is no longer just a future concept. Google and OpenAI now offer enterprise systems specifically designed around agents that connect to company tools, follow permissions and operate across multi-step workflows. Gartner forecasts that up to 40% of enterprise applications will contain task-specific agents by the end of 2026, compared with less than 5% when it made the forecast in 2025.

But there is an important reality check. Stanford’s 2026 AI Index found that actual agent deployment was still in the single digits across nearly all business functions during 2025.

So yes, agents are coming. No, you probably do not need an army of digital employees next Tuesday.

For SMEs, the best starting point is one well-defined process with a clear outcome and clear limits.

AI agents can take on routine tasks across email, CRM, scheduling and documents, helping teams focus on higher-value work.

2. The most useful AI will disappear into the software you already use

The biggest AI change may be the one you notice least.

Deloitte predicts that daily use of generative AI inside search engines will be three times more common than use of standalone GenAI tools in 2026 and beyond. More broadly, it expects much of AI’s growth to come through features added to software people already use.

That makes sense for a business.

Opening another AI website, copying information into it and then copying the result back into your CRM is not much of an automation. The bigger opportunity is AI working directly inside your email, CRM, finance system, help desk or document workflow.

For an SME, that could look like an incoming enquiry being classified automatically, a CRM record being updated, a draft response being prepared and the right colleague being asked for approval.

The AI itself becomes less interesting. The workflow becomes more important.

Rather than living in separate tools, AI will become embedded across everyday workflows, quietly assisting with tasks, insights and communication.

3. AI value will be found in boring, measurable work

This might be the least exciting prediction, but probably the most useful.

Stanford’s review of productivity research found some of the largest gains in structured, measurable tasks, including customer support, software development and marketing output. OpenAI’s own enterprise research similarly reports substantial numbers of users seeing faster marketing execution and IT issue resolution.

That points SMEs towards a simple rule: look for work that happens frequently, consumes measurable time and has an output you can check.

Good candidates might include first drafts of routine content, customer-service triage, meeting or document summaries, data classification, recurring reports, information retrieval, invoice processing or preparing information before a human makes a decision.

That does not mean every process should use AI. If a conventional workflow or automation rule can do the job more reliably and cheaply, use that instead.

AI should solve a problem, not become the problem.

The most valuable AI will connect directly with the systems businesses already use, bringing together data from CRM, finance, support and operations.

4. You may need your own AI workflow, but not your own AI model

Wise suggests businesses should consider building their own AI tools rather than relying entirely on generic products. The underlying point is sound, but for an SME we would interpret it differently.

Building your own AI does not have to mean training a giant model.

In many cases, the more practical approach is to use an established model while building your own business logic around it: your data sources, CRM records, approval rules, permissions, integrations and processes.

Current enterprise AI platforms are moving in exactly this direction. Google, for example, now supports multiple first-party and third-party models inside one agent platform, alongside orchestration, identity, security and evaluation tools.

That is useful because the “best” model will keep changing. Stanford reports that leading models are increasingly close on headline performance, shifting more of the practical decision towards reliability, cost and performance for a particular domain.

For most SMEs, your competitive advantage is unlikely to be the underlying language model. It is much more likely to be how well AI understands and fits your business process.

Businesses will increasingly judge AI by the results it delivers, from time saved and lower costs to better productivity and decision-making.

5. Human oversight becomes more important as AI gets more capable

The more an AI system can do, the more important its boundaries become.

An AI writing a first draft presents a very different level of risk from an AI that can modify a customer account, trigger a payment, send an external message or access confidential business data.

The UK government’s voluntary AI cyber security code calls for defined business requirements, security risk assessment, staff training, audit trails, controlled permissions, testing and human responsibility. EU requirements also impose human-oversight duties in specified high-risk uses.

This is why “human in the loop” needs to mean more than somebody glancing at an answer.

A sensible AI workflow should make it clear what the AI can access, what it is allowed to do automatically, what requires approval, what gets logged and when a person must take over.

That becomes particularly important around money, personal information, contractual commitments, security and regulated decisions.

As AI becomes more autonomous, human review, approval and oversight will remain essential for important decisions and actions.

What should small businesses do now?

Start with the process, not the AI product.

Find a repetitive task that genuinely costs the business time. Decide what a successful outcome would look like and how you will measure it. Work out what information the AI needs, what systems it needs access to and where a human approval step belongs.

Then run a controlled pilot.

If the result saves meaningful time without introducing unacceptable errors, security problems or additional administration, expand it. If it does not, stop.

That may sound less exciting than announcing an “AI transformation programme”, but it is much more likely to produce something useful. The available evidence increasingly suggests that organisational readiness, integration and workflow design matter at least as much as having access to a capable model.

Final thoughts

Wise is right about the overall direction: AI is moving deeper into everyday business processes, agents are becoming more capable, and people will increasingly supervise AI rather than simply ask it questions.

But for small businesses, 2026 does not need to be about chasing every new AI launch.

The opportunity is simpler: identify work that should take less time, connect the right technology to the right process, and keep sensible controls around it.

After all, the smartest use of AI is the one that genuinely makes your work, less work.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

A Day in the Life of an Operations Director

Mon, 02 Feb 2026 16:40:00 +0000

Holding a Growing Business Together (Before)

At 7:42am, Sam opens the laptop as the office begins to stir. The systems wake up quietly. Dashboards load. Inboxes refresh. Notifications arrive in quick succession. Sam takes a sip of coffee while scanning the overnight updates, already aware that the day will move faster than the information does. There are processes in place. There are systems for nearly everything. Yet the picture they create together is incomplete.

The first decision arrives before the morning meeting. A client escalation has been flagged overnight. Sam opens the CRM, then the service desk, then a shared document linked in an email. Each tells part of the story. Timelines do not quite align. Notes are missing context. The business needs an answer now, not later, so Sam makes the call based on experience and judgement. It is not guesswork. It is simply the best decision possible with what is visible.

By mid-morning, questions begin flowing in from different teams. Has that approval gone through. Are we still on track for delivery. Who owns the next step. Sam opens tabs almost instinctively. CRM. Project management. Email. A spreadsheet maintained carefully by one team because it fills gaps nothing else quite covers. The answers are there, scattered across systems that were never designed to work together. Sam replies with confidence, even when that confidence is supported by cross-checking rather than clarity.

Meetings follow. Good ones. Productive ones. Yet they circle the same issues. Visibility. Ownership. Handovers. Everyone is working hard. No one doubts that. The problem is not effort. It is friction. Small delays compound into uncertainty. Decisions slow as people wait for confirmation that should already exist.

Lunch happens between conversations. Sam approves invoices while responding to messages. Tasks are forwarded with notes asking for updates. People do their best to interpret what is needed. The organisation runs on professionalism and goodwill, supported by systems that require constant interpretation.

In the early afternoon, a customer raises a concern. Nothing dramatic. Just confusion. They were told one thing last week and are seeing another today. Sam listens, reassures, and promises to investigate. When the call ends, Sam already knows what will be found. No single error. No clear fault. Just multiple systems, each accurate in isolation, failing to present a single truth.

As the day moves on, planned work gives way to operational maintenance. Reports are assembled manually. Numbers are reconciled. Access needs adjusting for a role change that happened days ago. Sam stays focused, but the mental load builds quietly. This is not strategic work. This is compensating work.

At 6:30pm, the laptop finally closes. Sam reflects briefly before heading home. The business is performing. Clients are served. Targets are met. But it feels heavier than it should. Too much knowledge lives in people’s heads. Too many decisions depend on memory rather than visibility. The organisation functions, but only because experienced people are constantly holding it together.

Where Systems Carry Their Weight (After)

At 7:42am, Sam opens the laptop as the office comes to life. Dashboards load cleanly. Notifications are fewer, more meaningful. The day’s priorities are already visible. Context sits alongside meetings, tasks, and alerts. Sam takes a sip of coffee without rushing. There is time.

The first issue of the day appears clearly rather than urgently. A client escalation is flagged, already linked to the relevant project, support history, and recent communications. Sam reviews the information in one place. The timeline makes sense. The next step is suggested, based on similar cases and current capacity. Sam approves it and moves on. The decision is still human, but it is no longer made in partial darkness.

By mid-morning, questions arrive from teams, just as before. The difference is speed. Sam checks a single view. Status is clear. Ownership is visible. Updates are shared automatically. Answers take seconds. There is no need to cross-check or interpret. Everyone is working from the same version of events.

Meetings feel different too. Less time is spent aligning understanding. More time is spent improving outcomes. Data is trusted. Discussions move forward rather than looping back. Sam notices it without consciously trying to. The organisation feels calmer, not slower.

Lunch still happens between tasks, but the interruptions are fewer. Invoices are queued automatically. Role changes trigger access updates without reminders. Tasks appear where they should, already prioritised. Sam no longer needs to ask people to pick things up. The system has already done the coordination.

When a customer calls in the afternoon, the conversation is straightforward. Sam sees the full context instantly. What was agreed. What has been delivered. What comes next. The response is confident because it is informed. When the call ends, there is no follow-up scramble. The system already reflects the outcome.

Later in the day, Sam works on what the role was meant to focus on. Improving processes. Anticipating future demand. Making changes that reduce friction rather than reacting to it. Admin still exists, but it arrives in order and leaves quietly. It no longer competes with strategic thinking.

At 5:45pm, the laptop closes. Earlier than it used to. Sam does not think about what might have been missed. The systems are visible. The information is shared. The business does not rely on individuals to compensate for gaps anymore. People still matter deeply, but they are supported rather than stretched.

It is the same organisation. The same teams. The same ambition. The difference is not effort or talent. It is that the systems now carry their share of the load.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More

Business Technology Trends That Matter in 2026

Wed, 28 Jan 2026 08:00:00 +0000

In 2026, the organisations pulling ahead aren’t necessarily the ones buying the most technology - they’re the ones turning it into measurable outcomes: faster delivery, better customer experience, tighter security, and clearer decision-making.

That outcome gap is very real. PwC’s 2026 Global CEO Survey reports that only 12% of CEOs say AI has delivered both cost savings and revenue gains, while 56% say AI has delivered no significant cost or revenue benefit so far.

Below are the tech trends that matter most this year - with practical guidance throughout.

1. AI moves from “tooling” to operational decision support

Most businesses start with AI in a helpful-but-small way: content drafting, meeting summaries, basic assistance. Useful, but limited.

The organisations gaining advantage in 2026 are using AI for decision support inside workflows - where it influences prioritisation, next steps, and risk.

This trend is largely driven by a hard truth: many organisations invested in AI but haven’t seen returns yet. As noted above, PwC found 56% of CEOs report no significant cost or revenue benefit, and only 12% report both cost savings and revenue gains. PwC also reports CEO confidence about revenue growth is low: only three in ten (30%) CEOs are confident about revenue growth in 2026.

What this means in practice

AI creates the biggest impact where decisions are frequent and time-sensitive, for example:

Sales: next-best action, pipeline prioritisation, forecasting confidence
Support: ticket triage, suggested responses, escalation risk
Finance: anomaly detection, cashflow forecasting, spend pattern alerts
Ops: scheduling recommendations, capacity planning

The mistake many SMEs make is treating AI as a side tool. The win comes when AI is embedded into the systems teams already use (CRM, helpdesk, projects, finance) so that insight becomes action.

Actionable Takeaways

Start with one high-friction decision (e.g., “Which tickets should we handle first?” or “Which deals are most likely to close this month?”), then measure time saved and accuracy.
Create a simple ‘human-in-the-loop’ rule: AI suggests, humans approve (especially for customer-facing or financial outputs).
Treat data readiness as part of the AI project: if the CRM is inconsistent, AI outputs will be inconsistent too.
Use leading indicators (response time, backlog size, conversion rate) rather than vague success measures like “adoption”.

2. Automation becomes a margin-protection strategy, not just efficiency

In 2026, automation is less about being “faster” and more about being more consistent. SMEs are under pressure to deliver high service levels while controlling cost. Automation gives leverage when it removes:

Manual handovers
Duplicate data entry
Chasing updates
Inconsistent processes between staff members

This trend connects directly to the AI ROI story: businesses can’t get outcomes from AI if the underlying workflow is messy. Automation is often the bridge that turns “we know what to do” into “it actually happens every time”.

What this means in practice

The most valuable automations aren’t flashy - they’re the ones that eliminate repeatable admin work and reduce errors:

Sales won → create onboarding tasks + schedule implementation steps
Support ticket created → categorise + assign + set SLA + notify customer
Invoice overdue → trigger reminder sequence + internal alert
Offboarding → revoke access + archive data + confirm handover steps

Actionable Takeaways

Map one end-to-end journey (lead → sale → delivery → invoice → support). Most inefficiency sits in the gaps.
Automate the handoffs first (handoffs create delays and errors).
Automate “status movement” (changes in stage should trigger tasks, reminders, and checks).
Monitor outcomes, not just time saved: fewer escalations, faster cash collection, fewer missed steps.

3. Cybersecurity becomes business continuity, not an IT topic

Security incidents are now normal business risk. The UK Government’s Cyber Security Breaches Survey 2025 reports that 43% of UK businesses experienced a cyber breach or attack in the past 12 months.

What this means in practice

Among those that were breached/attacked, phishing was the most prevalent type by far - experienced by 85% of affected businesses (and also described as among the most disruptive).

But preparedness lags. A UK Government visual summary drawing on the same survey notes that only around 1 in 5 businesses (19%) provided staff cyber security training in the prior 12 months.

What this means in practice

The biggest risks for SMEs are often not advanced hacking - they’re:

Compromised email accounts
Weak identity controls
Staff clicking convincing phishing emails
Slow response when something goes wrong

This is why security is becoming a competitive advantage: customers increasingly choose vendors they trust to protect data and maintain continuity.

Actionable Takeaways

Make identity the perimeter: enforce MFA everywhere, remove shared accounts, review admin access.
Reduce your “blast radius”: least privilege access (people should only access what they genuinely need).
Practise incident response: define what happens in the first hour of an incident (who does what, how you contain, how you communicate).
Run short training regularly: 10 minutes monthly beats an annual checkbox session (and tackles the phishing reality in the survey).

4. Integration beats tool sprawl - complexity is the hidden productivity cost

Businesses are using more apps than ever - and that complexity is expensive. Okta’s 2025 Businesses at Work report states that the global average number of apps per customer has topped 100 for the first time, driven by 9% year-on-year growth.

More tools usually means more silos. Salesforce reports 81% of IT leaders say data silos are hindering digital transformation.

What this means in practice

Tool sprawl creates invisible drag:

People can’t find the latest information
Teams duplicate work across systems
Onboarding takes longer
Reporting becomes unreliable

Integration is not about having fewer tools - it’s about having clear systems of record and reliable flow of data between them.

Actionable Takeaways

Define your systems of record (one place that is “truth” for customers, billing, tickets, projects).
Eliminate manual copying: if staff copy/paste between tools, that’s a high-value integration target.
Build dashboards off live data (not exported spreadsheets).
Retire low-adoption apps: every extra system increases support burden and security exposure.

5. Data quality becomes the limiter for growth and AI

As organisations push more automation and AI into operations, data quality becomes non-negotiable. Gartner states that poor data quality costs organisations at least $12.9 million a year on average.

What this means in practice

Even if you’re not a giant enterprise, the pattern holds: poor data causes:

Wrong decisions (bad forecasting, misprioritisation)
Wasted time (cleanup, reconciliation, “which number is right?”)
Lost sales (duplicate or incomplete CRM records)
Weak AI outputs (because AI relies on the same underlying data)

If people don’t trust your dashboards, they’ll rebuild spreadsheets. Once that happens, you lose a shared reality - and scale gets harder.

Actionable Takeaways

Assign data ownership (CRM owner, finance owner, support owner - named people).
Standardise key fields that drive decisions (lifecycle stages, industry, renewal date, lead source).
Use controlled inputs: dropdowns and validation rules beat free text.
Schedule monthly hygiene (dedupe + incomplete records + incorrect statuses).

A simple 60-day plan most SMEs can actually execute

AI works best when it’s built on strong foundations, not dropped onto messy processes.

Weeks 1 – 2

Choose one measurable outcome (e.g., “reduce support backlog by 30%”) and audit the workflow + data feeding it.

Weeks 3 – 4

Implement automation at handoffs + tidy the key data fields.

Weeks 5 – 8

Add AI for decision support (triage, recommendations, summaries) with human review, then measure impact against your baseline.

Weeks 9 – 12

Expand to the next workflow only if the first one delivered measurable improvement.

The defining difference between businesses that thrive in 2026 and those that struggle won’t be how much technology they buy - it will be how deliberately they use it.

The data is clear:

AI only delivers value when it’s embedded into real workflows, not run as an experiment.
Automation protects margins when it removes friction between teams, not just speeds up tasks.
Cybersecurity is now a core business risk, driven largely by identity and human factors.
Too many disconnected tools quietly destroy productivity - integration matters more than ever.
Poor data quality undermines every investment that sits on top of it, from reporting to AI.

For most organisations, progress doesn’t come from wholesale transformation. It comes from small, well-chosen improvements that compound over time: fixing one workflow, tightening one dataset, or solving one recurring decision properly.

The most successful businesses in 2026 will be the ones that stop asking “What technology should we adopt next?” and start asking “What’s slowing us down - and how do we fix it for good?”

That shift in mindset is where technology stops being a cost centre and starts becoming a strategic advantage.

Contact

Get in touch with the team to discuss how we support your business with practical, people-first technology and long-term solutions.

About

Learn who we are, what we stand for, and how Ostratto helps businesses make their work, less work through practical technology solutions.

Learn More

Our Approach

Discover how we partner with you - focusing on strategy, simplicity and long-term value to deliver technology that truly supports your business.

Learn More