Wiki › Work, teams and companies

AI Adoption Mandates

Top-down pushes that make AI use itself the goal through usage tracking, job leverage and manifestos, and can silence security and engineering feedback.

An AI adoption mandate is a top-down push in which leadership sets AI use itself as the goal. Teams are measured on how much they use AI and how fast they ship, and the quality of the results matters less. The best-documented case so far is Proton in 2026.¹

Typical levers

From the Proton reporting:¹

• Measuring usage. Software on employees' machines tracks the share of AI-written code. In October, 90% of Proton's code was AI-written.
• Removing friction. A dedicated address collects reports of anything blocking "AI velocity" so leadership can "ruthlessly smash" it.
• Job leverage. Staff are re-interviewed about their AI use, and leadership says it already knows "who is behind". Proton also announced that it would stop hiring engineers and hire AI-assisted product builders instead.
• Manifestos. The CTO's Friday-afternoon "your old job is gone" document was meant to provoke. It set the goal that most code be reviewed by AI rather than people.
• Calling caution weakness. The CEO called pentesting before release "loser thinking" and told engineers to "prompt better".

Failure modes

• Safety work gets cut. One Proton engineer said "security is low in our prioritization at present". If speed is the metric, testing and review read as obstacles. In a privacy product that conflicts directly with the brand promise and with the defensive needs described in AI Vulnerability Tsunami.
• Silence. Employees describe being afraid to raise engineering problems in public channels. A mandate enforced by fear removes the feedback loop it needs. One senior engineer called a fear-based culture "the opposite of engineering".
• Usage replaces outcomes. Counting AI-written lines measures adoption, not working software. This is the organisational version of Cognitive Surrender and the Speedup Illusion. Engineers also doubted that agents can maintain a large codebase over time, which is the Loop-Dependent Software risk.
• Mixed motives. Proton tied "insufficient velocity" to a missed revenue target while denying a financial motive. That makes it hard to tell technical conviction from cost pressure.

Contrast

Other organisations spread AI by making it visible and easy to learn, as in Public Agent Collaboration around Shopify's River. Felix Rieseberg's Newcomer Advantage rests on earned influence through prototypes, not surveillance. Teams that do drop line-by-line review, as in Agentic Software Factories, usually build strong automated validation first. At Proton the review-free goal arrived before any comparable verification layer had been described.

Caveats

This rests on one investigative report built from anonymous employees and leaked material, though Proton did not dispute it. It covers one company. Whether mandates generally speed up useful adoption or mostly produce compliance theatre is an open question for AI Labor Market and AI Usage Telemetry.