Adoption hit 85.8%, but the rule about not shipping model output was settled first

The measures companies picked sit at the entrance, the middle and the exit

The Computer Entertainment Supplier’s Association (CESA) published the preview edition of the CESA Games White Paper 2026 on 17 September 2026. The report carries two surveys CESA ran in 2026, both appearing for the first time. One went to game developers, the other to CESA member companies.

Asked in the member survey how they manage generative AI use, the answer that drew the most responses was that a person checks, corrects and reviews the output. Restricting or designating which tools may be used, and not shipping model output as-is, also drew many responses.

Those three answers point at three different moments in the work. Designating tools narrows the entrance. The principle of not shipping output as-is operates mid-process. Human review is the last gate. A person stands at the entrance, in the middle and at the exit. Controlling all three at once is a signal that these companies treat adoption and trust as separate decisions. They have settled on using it; they have not settled on believing what comes out.

The developer numbers say the adoption argument is over

In the same report’s developer survey, 85.8% of respondents said they use generative AI in their work. Inside that, 63.0% said they use it routinely and 22.8% said they use it occasionally. Another 8.6% said they are trialling it for evaluation, 4.8% said they have never used it and 0.7% said they used to but no longer do.

The gap worth noting is the one between 63.0% and 8.6%. More than seven times as many people already reach for it every day as are still deciding whether it works. The adoption question is settled on the floor: it was open yesterday and it is open today. Only 0.7% stopped. That almost nobody tried it and walked away is not the same as everybody being happy with it. A tool that has entered a workflow is hard to pull back out.

Laid side by side the two surveys look contradictory, but they sit on different layers. The 85.8% came from developers describing their own work; the expected benefits and management measures came from companies answering as employers. Developers using it daily and companies deciding not to ship its output are both true at once.

When the goal is efficiency, review time lands on the other side of the ledger

Asked what they expect most from generative AI, member companies named efficiency and higher productivity. Not making something new, and not doing what people could not do before, but finishing existing work faster.

When that is the goal, the question left after adoption changes too. It is no longer what can be made, but whether what was made at this speed can go out as it is. The yardstick moves with it. Instead of what is new, the measure becomes how many days the same volume took. On that yardstick, review time is a cost line.

Here is the trap that is easy to walk into. Buying a tool ends with a contract, but keeping a review rule spends people’s hours forever. Measure the benefit only in drafting time and the reviewing hours that grew underneath never reach the books. How tightly the review rules are drawn is therefore bound up with what the adoption is worth.

Three lines to settle on a single page

What this survey hands a founder is not a trend but three sentences. The answers the member companies picked become the three items. First, list the tools that may be used and write down what happens to anything outside the list. Second, write down which deliverables the no-raw-output principle covers. Third, name who looks last and what line that person cannot let through.

The value of writing those three lines early shows up later. A team of five can agree out loud, but the moment contractors and fixed-term staff arrive the same question comes back. With no document, people answer it differently and the work has already shipped. What the report shows is that companies put that ordering ahead of adoption itself.

One thing the preview announcement does not settle. Which stage of production the check-correct-review answer refers to, and whether tool designation is an internal policy or a clause in an outsourcing contract, cannot be told from what is public now. The full edition arrives in early December. Which of the two it is decides whether the same sentence belongs in the internal wiki or in an annex to the contract.