A patch can be written in minutes and still spend the afternoon waiting for a build, a decision or someone to review it. Proton’s newly published engineering manifesto puts that waiting inside the AI adoption problem.

In his November 1 article, CTO Bart Butler publishes an internal memo written about a month earlier. He identifies three obstacles: engineering habits, workflows built around existing constraints, and insufficient shared investment. The memo asks engineers to guide agents, improve reusable instructions and address recurring failures. It also argues that faster code production can move the constraint into testing, specifications, approvals or review.

This is an employer’s proposed direction. The article supplies no controlled productivity comparison or technical plan for reliability and security. Teams adopting its direction would still need to discover where their work now stops.

For a delivery team, that suggests following one change from request to accepted release. Record how much time goes into drafting, waiting, checking and repairing. If an agent saves an hour of coding while the build queue adds two, the bottleneck has moved. Counting generated changes alone would miss it.

There is also a question of ownership. An engineer may be able to improve an instruction file immediately. Repairing a shared test environment may require another team’s time and a manager’s budget. Giving the engineer responsibility for the whole result needs a corresponding route to get that work funded.

A useful pilot would make one recurring failure cheaper for the next person, then check whether that improvement survives another task. The evidence to bring to a planning meeting is a shorter path to an accepted change, with its repair burden still visible.