In the 2027 budget review, the question about the development team's AI will no longer be whether it is used. It will be what that money has bought.
This piece argues a thesis that makes anyone who has spent a year counting licenses uncomfortable: with AI at 17% of the IT budget, the adoption rate is no longer an argument. Each development team needs its own value thesis (savings, growth or a staged bet) and the evidence that goes with it. What to measure comes after what you want to prove.
The figure: from 5% to 17%
On 2 October 2026, BCG published an analysis of how AI is changing the economics of technology in companies. Its central figure comes from the Applied AI 2026 survey: AI's share of the IT budget has more than tripled, from 5% to 17%. And that is not all of it: BCG estimates that companies spend on average 3.3% of revenue on enterprise IT plus a further 2.8% on AI initiatives funded outside that budget.
The page does not publish the sample size, so the figure is best read as an order of magnitude. Even so, what matters is the change of scale: a 5% line can be treated as an experiment; a 17% line enters the conversation with the CFO, and that conversation runs on business arguments.
Why the adoption rate no longer works
"Almost the whole team uses an assistant every day" describes activity, not results. BCG says so for the whole company: it recommends measuring cost per outcome rather than activity, and warns that CIOs should not view AI as a cost-reduction program. What sustains the investment is turning it into productivity that can be banked, into revenue growth, or both.
In a development team there is one more reason. DORA, in its 2025 report on AI-assisted software development, concludes that AI's primary role is as an amplifier, magnifying an organization's existing strengths and weaknesses. The same adoption can speed up delivery in a team with good tests and good review, and multiply rework in one without them. The usage rate does not tell the two apart.
Three theses, three kinds of evidence
The most useful part of BCG's analysis is one sentence: every scaled AI investment needs an explicit value thesis and evidence appropriate to its purpose, which can be banked efficiency, measurable growth or staged strategic value. And if it repeatedly fails that test, it should be rescoped or stopped. Applied to a development team, it looks like this:
| Thesis | What it promises in a development team | Evidence that goes with it | Signal to rescope |
|---|---|---|---|
| Banked efficiency | The same work at lower cost, with the saving taken out of a budget line | Cost per delivery against an earlier baseline, with quality the same or better | The saving does not show up in any budget line |
| Measurable growth | More product in production: features, customers served, less time to production | Deliveries reaching production and their business effect, against their baseline | More code and the same delivery |
| Staged strategic value | A capability that does not exist today, such as maintaining a legacy system nobody on the team understands | Milestones with dates and a criterion for moving to the next one | Milestones slip without anyone deciding so |
Prepared by onext from BCG's three categories (2026); the examples are illustrative
A value thesis is not an adoption target. It is a sentence that can be proven false, with a date: "this team will deliver the same at lower cost before June" or "by the second quarter we will know whether we can maintain this system without its original author".
The third row is not the most comfortable, but it is often the most honest. If nobody on the team can explain part of the code, the value of AI may lie in recovering that understanding, as we explain in comprehension debt, and that is not measured in hours saved.
What the KPMG figure says, and what it does not
In September 2026 KPMG published its Global AI Pulse for the third quarter, with 2,131 senior leaders from 20 countries and territories and fieldwork between 23 July and 26 August. Among organizations reporting established ROI, 86% operate a formal cross-functional or enterprise-wide AI management layer. Among those at the experimentation stage, the share is 31%.
Risk This is an association, not proof of cause. Governing AI may help to obtain a return, and it may be that only organizations already making money from AI can afford to govern it. What the figure does suggest is that those who prove a return have someone who decides what is measured and when it stops.
The tool amplifies what is already there
If AI amplifies, a team's thesis depends on what AI is going to amplify. An efficiency thesis in a team without automated tests or review ends up saving on writing and spending on fixing. That is why, before promising savings, it is worth having the net that catches mistakes, as we explain in before the agent, the safety net. DORA also publishes a model of seven capabilities that amplify the benefits of AI: a good starting point for reviewing what the team has before writing the thesis.
And if the adoption problem is organizational rather than a matter of tooling, changing tools will not fix it, as we explain in the 70% gap.
Writing the team's thesis on one page
The thesis fits on one page and is written before anything changes. This is the order:
- Choose a single thesis for the team: efficiency, growth or staged value. If you say all three, you have said none.
- Write the sentence that could be proven false, with a date and with what has to happen.
- Set the baseline before touching how the team works. Without the before, the after means nothing.
- Choose the metrics that go with that thesis and only those. The others are watched, but they do not decide.
- Agree with finance which result forces a rescope or a stop, and when it is reviewed.
For the third step, how to build the baseline for a first case is in choosing and measuring your first use case. For the fourth, the list of possible metrics is in the KPIs of a development team with AI: the thesis decides which of them count. And why counting active licenses is not measuring return is explained in the ROI of Copilot and Cursor.
How we approach it at onext
If you want to write your team's thesis with someone from outside, the diagnostic is free, and if the diagnostic shows no difference, there is no Quick-start. And we do not publish productivity or speed figures that we have not measured with a verifiable source. The rest of the way is described in AI-Accelerated Development.
Frequently asked questions
How much of the IT budget does AI take up?
According to BCG's Applied AI 2026 survey, published in October 2026, AI's share of the IT budget has more than tripled, from 5% to 17%. BCG adds that companies spend on average a further 2.8% of revenue on AI initiatives funded outside the IT budget. The page does not publish the sample size, so it is best read as an order of magnitude.
What is a value thesis for an AI investment?
It is what the investment has to prove, stated in a way that can be checked. BCG distinguishes three: banked efficiency (savings already taken out of the budget), measurable growth and staged strategic value. Each calls for a different kind of evidence, and if the investment repeatedly fails that test, BCG recommends rescoping or stopping it.
Why is measuring AI adoption in the team not enough?
Because the share of developers using an assistant describes activity, not results. DORA concludes that AI acts as an amplifier of the strengths and weaknesses an organization already has: the same adoption can speed up delivery in one team and multiply rework in another.
Which thesis fits a development team?
One per team, matching what the business expects from it. Efficiency, if the goal is to deliver the same at lower cost and that saving will come out of a budget line. Growth, if more product is expected to reach production. Staged value, if the aim is to build a capability that does not exist today, with milestones and a date.
Does formal AI governance guarantee a return?
No. KPMG finds that 86% of organizations reporting established ROI operate a formal AI management layer, compared with 31% of those still experimenting. That is an association, not proof of cause: it may also be that only organizations already making money from AI can afford to govern it.
When should an AI investment in a development team be stopped?
When it repeatedly fails the test of its own thesis: the saving does not show up in any budget line, more code arrives but not more delivery, or milestones slip without anyone deciding so. That criterion is agreed with finance before starting, not when the budget review arrives.
Sources
- Michael Grebe, Joppe Bijlsma, Djon Kleine et al., "How AI Is Reshaping the Economics of Technology", BCG, 2 October 2026 (Applied AI 2026 survey; sample size not published on the page).
- KPMG, "Global AI Pulse Q3 2026", September 2026 (2,131 senior leaders from 20 countries and territories; fieldwork from 23 July to 26 August 2026).
- DORA, "State of AI-assisted Software Development 2025", 2025 (no exact date visible; accessed 19 October 2026).
- DORA, "DORA AI Capabilities Model", updated 25 November 2025.
- The table of theses, the examples (illustrative, with no client behind them) and the steps were prepared by onext.
Bernat López is founder and CEO of onext, an AI boutique. He helps development and product teams work with AI with method —specification before building, a person who decides where there is risk and Spec-Driven Development— and applies to his own company what he proposes: onext runs on its own agentic system.
LinkedIn →