AI and productivity: where the gains show up, and where they don't

Stanford HAI's AI Index Report 2026 compiles, in its chapter on the economy, a dozen independent studies on AI's actual effect on productivity – both at the task level and in aggregate. It's a more useful picture than a single adoption or expectation figure, because it shows where the gains actually show up, and where they still don't.

Where the gains are already measurable

At the task level, the pattern is clear in a handful of specific occupations. Customer support agents using a conversational AI assistant resolved 14-15 percent more issues per hour (Brynjolfsson et al., 2025). Developers using GitHub Copilot completed 26 percent more pull requests (Cui et al., 2025). Marketing teams using AI for ad creation saw a 50 percent increase in output per worker (Ju and Aral, 2025).

What the three have in common is that the work can be broken down into well-defined, repeatable tasks with clear quality checks – exactly the kind of work where, according to the report, the gains are strongest.

The biggest winners are rarely the experts

A recurring finding across several of the studies is that less experienced workers benefited the most from AI support, suggesting the tools can help close skill gaps within a team rather than simply making the already skilled even faster. That's a concrete argument for giving AI to the whole team, not just the most senior people.

Where deeper judgement is required, the picture changes

The picture isn't uniformly positive. For work that requires deeper reasoning, a widely cited study from Model Evaluation & Threat Research (METR) found that experienced developers became 19 percent slower when using AI assistance, with a clear gap between how helpful they believed the tools were and how they actually performed (Becker et al., 2025). METR has since been unable to replicate the result, partly because fewer developers are now willing to work without AI at all. But the underlying conclusion holds: AI's productivity effect is highly context-dependent, and the gains are strongest where work can be divided into well-defined, repeatable tasks with clear quality monitoring.

At the macro level, the picture is still early

A level up, at the firm and economy scale, the evidence is thinner and more mixed. A study of 12,000 European firms found that AI adoption boosted labour productivity by 4 percent, with training strengthening the outcome further (Aldasoro et al., 2026). OECD projections for G7 economies point to annual productivity gains of 0.2 to 1.3 percentage points over the next decade (Filippucci et al., 2025). At the same time, a survey of 6,000 executives across four countries found widespread adoption but minimal realised productivity gains so far, and a projected 0.7 percent reduction in employment over the next three years (Yotzov et al., 2026).

The report's own summary puts it well: the gains are measurable within narrow tasks, but the evidence at the macro level remains early and mixed.

What it means for where you start

This matches how we already think about where AI does the most good: look for work that is repetitive and pattern-based, with a clear definition of what a good result looks like, rather than rare judgement calls. That's not a compromise – it's where the report's own numbers show the gains actually materialise. And since it's often the less experienced people who benefit most, there's rarely a good reason to limit AI to the most senior members of a team.

Where Nodal comes in

We help companies find exactly that kind of task in their own operations – well-defined, measurable, with a clear quality check – and build AI support around them, instead of guessing where AI ought to make a difference.

Wondering where in your business AI would actually deliver measurable value? Get in touch with us at Nodal, and we'll go through it together.