Proton will re-interview staff on their AI use, and the engineers asking how AI code gets checked are doing the job it says it needs

404 Media reported this week that Proton, the Swiss company behind the encrypted email service Proton Mail, will re-interview staff about how they use AI in their jobs. Employees read the interviews as deciding who stays. "These re-interviews are consequential," the company's email said. 404 Media spoke to eight current employees and one former employee. It also reviewed Slack chats, meeting recordings and an AI manifesto written by chief technology officer Bart Butler. A Proton spokesperson did not dispute any of the reporting.

What Proton is counting

Proton installed software on workers' computers that tracks how much AI code the company generates. Butler said in a meeting that 90% of Proton's code in October was written by AI, according to audio 404 Media obtained. Chief executive Andy Yen set up an email address so leadership can "ruthlessly smash everything that stands in the way of AI velocity". He also said this is the last year Proton will hire engineers.

The headline misses the reason behind all this. An internal document seen by 404 Media says that "due to insufficient velocity" Proton is not on track to hit its annual recurring revenue target by the end of the year. It adds: "By all metrics that matter, we're an underperforming company." Proton's AI push is a revenue plan, and the number it watches is volume.

The cost shows up in security. At the end of August an engineer asked for a few days to pentest new code before it went live. Yen called that "bullshit and loser thinking". Another engineer wrote: "AI won't magically solve our problems, because security is low in our prioritization at present." Butler's manifesto calls for a Proton where most code is reviewed by AI rather than by humans.

This is a story about who signs off. When a company measures how much code AI produces, it can no longer see the person who knows whether that code is safe to ship. Yet that person carries the company's risk.

The role Yen describes is judgement

A senior employee suggested that Proton find the people who were falling behind and train them. Yen replied: "we already know more or less who is behind. If we were giving up on them, we would not bother with this process."

Ignore the tone and look at Yen's own forecast. He wrote that by the end of this year most developers will "start writing specs for tests that AI will code and run". From next year, he said, Proton will hire "what we used to call product managers, to build things with AI." The job he describes is to say what should be built, define how to check it and answer for the result.

AI does the routine work. You do the thinking. The engineers asking how AI-written code gets verified were already doing the thinking. Their mistake was asking a question without bringing a plan to answer it at speed.

Your Next Move

1. Keep a record of outcomes from today. For each piece of AI-assisted work, write down what shipped, what you checked and what you caught before release. A tracker like Proton's can count usage. Only you can show impact. Our guide to showing impact when your employer grades your AI use sets out how to present that record before your next review.
2. Raise a risk together with a fix. Andy Yen called the Proton engineer's request for days of pentesting "loser thinking" in the company's security Slack channel. Propose a check that keeps pace with the team, such as an AI-assisted review step with a named person who signs it off. That makes you the person who keeps speed safe.
3. Write one spec this week. Pick a task you would normally build by hand. Write down what it must do and how you would test it, then let a model produce the first draft. Yen says this is the skill Proton will hire for, so practise it now.