AI Spilled Engineering Skill Into Every Occupation and Design Skill Into Almost None — and in OpenAI's Crossover Data, the Next-Month Return Rate Is Only 18.5%

OpenAI's Economic Research team published the second installment of its Work at the Frontier series, "How workers are unlocking new ways of working," analyzing more than 1.5 million work-related ChatGPT messages from April through July 2026. The core finding is that cross-occupation AI use recurs: workers return to tasks outside their own occupation, and those activities grow as a share of their observed AI use over time — the mechanism being to try something outside your role once, find AI useful for it, and then make it part of the routine. But the recurrence rate is modest: the average next-month return rate across cross-occupation tasks is 18.5%, varying by task — around 15% for explaining financial information — with researchers suggesting the differences may reflect whether AI fits naturally into a recurring workflow, or differences in workplace norms, caution, and the consequences of errors. The first installment (July, by Caroline Chin and Alex Martin Richmond) analyzed more than 800,000 messages from US users: 16.8% of work-related messages and 43.5% of occupation-specific messages involved tasks associated with another occupation. The direction is strikingly asymmetric — about 35.2% of designers' messages reached into other occupations, while design tasks made up only 1.7% of everyone else's messages; engineering was the reverse, with only 18.5% reaching outward but engineering tasks accounting for 7.4% of messages from other occupations. The pattern is strongest at smaller organizations.

The Asymmetry Is More Interesting Than the Totals

Put two numbers together: engineering tasks account for 7.4% of other workers' messages, design tasks for only 1.7%. AI spilled engineering capability across every occupation and barely spilled design capability at all. That is not because design is harder. The likelier explanation is a difference in **verification cost**: whether a piece of code runs is answered immediately; whether a design is good has no compiler. The skills AI spreads most easily are the ones with fast, objective, cheap feedback loops — writing a script, chasing an error, composing a regex, fixing a config, all of which tell you instantly whether you got it right. Design taste, writing voice, negotiation: even when the model can produce them, the person using it cannot judge whether to accept the output. That inference is directly usable: **which field you can safely cross into with AI depends on whether right and wrong in that field can be verified by you, on the spot.** Where it can, cross freely. Where it cannot, the risk of crossing is that you won't see the mistake. The numbers in the other direction say something too: designers reach outward in 35.2% of messages, sales 32.1%, HR and customer experience 30.4%. The occupations reaching out most are exactly the ones whose daily work touches technical matters without having technical people on hand.

18.5% Is a Cooling Number

The figure to remember from the second installment is that return rate. The average next-month return rate across cross-occupation tasks is 18.5% — meaning most crossover attempts are one-offs, and fewer than one in five become habits. Against the claim that AI lets anyone do every job, that is a plain counterweight. The cost of trying has indeed fallen to nearly zero, but **turning one attempt into standing work still has to clear process, norms and accountability.** The researchers' own explanation points the same way: the variation may come from whether AI fits naturally into a recurring workflow, or from workplace norms, caution, and the consequences of errors. One conclusion from the first installment is worth keeping alongside it: crossover does not replace role-specific work, and every occupation retains a meaningful share of same-occupation activity. What is visible is not occupational boundaries disappearing but boundaries becoming elastic.

Smaller Organizations Benefit Most, Which Is the Most Practical Finding

The first report found the pattern strongest at smaller organizations — no specialist readily available, so the worker does it themselves with AI. That is useful for both tooling decisions and headcount planning. On a small team with no dedicated designer, lawyer or data analyst, AI fills exactly that vacancy and the return is most direct. In a large organization where the specialist sits one desk over, the marginal gain from crossover is far smaller, because the original cost was never "nobody can do this" but "waiting for them." In other words, **the same tool substitutes for entirely different things depending on organization size**: at a small company it substitutes for missing capability, at a large one for coordination delay. The boundaries need stating: all the data comes from ChatGPT's own messages, covering only its users and behavior it can observe; occupational classification is made by OpenAI's classifier; and the first installment is limited to US users. This is vendor-owned data on vendor-defined terms, with no external reproduction. The series also builds on OpenAI's April AI Jobs Transition Framework, which predicted 24% of US jobs are likely to "reorganize" as their day-to-day tasks shift — that is a forecast, not an observation, and the two should not be read together as one.

via: OpenAI: How workers are unlocking new ways of working, OpenAI: How AI is expanding what people do at work, the Work at the Frontier report PDF