Is AI a fad? Why it's becoming the operating model.
The serious money has stopped buying software and started buying finished work. Here is what that shift looks like in the data, and what it changes about the job you actually do.
Every founder pitched AI in the last 18 months has the same question. Bubble or structural shift? The numbers from the major research houses now answer it clearly enough to act on.
Is AI a fad?
No, on the evidence so far. Regular use is now reported across most surveyed organisations, while scaled agentic systems remain rare, so the shift is real but uneven. The advantage is going to people who move from occasionally using AI to running their work with it.
Is AI in business a fad or a structural shift?
It is a structural shift, and the gap between using AI and running on it is the new arbitrage. McKinsey's State of AI 2025, published November 2025, set the baseline. Adoption is no longer a curiosity in a Friday-afternoon experiment slot. It is sitting on the operating P&L.
Sequoia Capital's "2026: This is AGI" essay argues that long-horizon agents make it possible to sell completed work rather than access to software. That is an investor thesis, not a measured market outcome, but the commercial distinction is useful: a tool helps a bookkeeper; an outcome-based service closes the books.
The Sequoia thesis in one sentence
Buyers stop paying for access and start paying for outcomes. Books closed. Contracts reviewed. Claims handled. Returns reconciled. The product is the job done, not the dashboard you log into to do the job. That re-prices a large slice of the mid-market operations stack.
The buyer wanted more than the software. They wanted the close.
If you have watched a finance team queue up another month-end ritual, you can see what this could re-price. The software licence and the labour needed to operate it become one service line priced against an agreed output. Whether it is cheaper must be tested against the real workflow.
The agentic adoption gap
Use is widespread, while scale remains early. McKinsey's State of AI in 2025 reports 88% regular use in at least one business function and 23% scaling an agentic AI system somewhere in the enterprise. BCG's AI Radar 2026 reports nearly all surveyed CEOs expect agents to produce measurable returns in 2026. High expectation is not proof of realised value.
Deloitte's State of AI in the Enterprise 2026 identifies customer support as the area where respondents expect agentic AI to have the highest impact, with supply chain and knowledge management also prominent. Candidate work includes month-end reconciliation, invoice processing, ticket triage and supplier onboarding, but expected impact is not the same as independently measured savings.
Dead SaaS is the new arbitrage
Whole categories of SaaS now look strandable. Products built on the assumption a human would do the workflow behind the dashboard are being out-competed by agents that ship the result and skip the dashboard. The pattern is visible across invoice extraction, expense parsing, e-commerce reporting and ad-account audits. The previous answer was a tool plus a person. The new answer is the outcome, on a usage price. When the work is done by an agent, the dashboard is the audit trail, not the product.
What this means for the work you do
UK adoption is uneven. DSIT's AI Adoption Research, published in January 2026, found 16% of surveyed UK businesses with at least five employees were using one or more AI technologies. Large businesses reported higher use than micro businesses. That creates an opportunity, but not a reason to skip workflow selection, controls or measurement. Where projects stall is often organisational.
Subscription stacks become operations stacks.
Reconciliation, invoice processing, supplier onboarding and marketing reporting move from SaaS line items to agent line items, priced on outcomes.
The next ops hire is a choice, not a default.
If a role is mostly repeatable rules and structured inputs, an agent will be cheaper and faster. See the headcount maths. Judgement and exception handling still need people.
Data and controls decide readiness.
The source fields, definitions, access rights and exception history must be assessed workflow by workflow before an agent is trusted with live work.
What this isn't
It is not "fire everyone". The functions where agents work reliably today are narrow and rules-based. Work that needs context, taste or political judgement is not going anywhere. The shape that wins is small teams running agents, not no teams.
It is not "deploy ChatGPT and you're done". A team using ChatGPT casually is not the same as an agent wired into your data, running a workflow end to end and producing auditable output every time. The first is a productivity nudge. The second is an operating-model change, and only the second shows up on the cost line. Both start in the same place, though: one person getting reliably useful results out of an assistant, which is exactly what our practical starting point for AI at work teaches.
So why is this generational
Because the unit of sale can move from access to outcome, and the unit of cost can move from fixed labour towards variable consumption. The opportunity for a UK operator is to test that model on a bounded workflow, not to assume every dashboard or role can be replaced.
AI is moving from experiment towards operating model, but scaled value is not yet universal. Audit repetitive workflows before the next hire, establish the baseline and controls, and buy measurable outcomes rather than another unsupported promise.
Frequently asked questions
Is AI in business a fad or a structural shift?
AI use is now widespread enough to affect operating models, although enterprise-scale value remains uncommon. McKinsey's 2025 survey reports 88% regular use in at least one function and only 23% scaling an agentic system somewhere in the enterprise.
What is the agentic adoption gap?
It is the distance between using AI and scaling controlled agentic systems. McKinsey's State of AI in 2025 reports 88% regular AI use in at least one function, while 23% report scaling an agentic AI system somewhere in the enterprise.
Which business functions are candidates for AI agents?
Deloitte's 2026 research identifies customer support as the area where respondents expect the highest impact from agentic AI, with supply chain and knowledge management also prominent. Finance and operations workflows still need individual assessment for data, controls and human review.
What should a UK founder do first?
Audit repetitive workflows before selecting a tool. Establish the baseline, data access, exception rules, control owner and success measure for each candidate. Then compare the build with the cost of the current workflow.
References
McKinsey, The state of AI in 2025; Sequoia Capital, 2026: This is AGI; BCG, AI Radar 2026; Deloitte, State of AI in the Enterprise 2026; and Department for Science, Innovation and Technology, AI Adoption Research.
The UK adoption figures quoted above, with their denominators, sample bases and caveats set out in full, are collected in our roundup of UK AI adoption statistics.
Put AI to work in your own job.
AI at Work is Clerq's practical, hype-free guide to getting real work done with AI - written for people with inboxes, deadlines and meetings, not developers. PDF and EPUB, launching soon.