AI mandates are the easy part
CEOs are mandating AI adoption in engineering teams, with real consequences for opting out. Writing the memo is the easy part. What separates the teams that get faster from the teams that only get noisier is who owns the practice afterward.
In 2025, the CEO of a major US exchange gave his engineering organization a deadline: onboard to the company's AI coding tools by the end of the week. Engineers who had no good reason for skipping it were let go. The story was covered widely because of the firings, but the more useful detail is what the CEO pushed back on: his teams had proposed reaching 50 percent adoption over a quarter, and he refused the timeline, not the goal.
That is where most engineering organizations are in 2026. The debate about whether to adopt AI is over; boards and CEOs have closed it. Engineering leaders now write some version of the same line one team lead put publicly: using AI is no longer optional. And many of them admit, in the same breath, that the mandate makes them uneasy.
Why mandates stall at the adoption number
A mandate produces onboarding. It does not produce practice. The gap shows up in three places:
- Adoption is measured, impact is not. Teams report percent adoption because it is countable. Engineers on public forums point out the obvious problem: tool-use mandates arrive without anyone quantifying what the tools change in delivery. When the metric is logins, the result is logins.
- Usage is uneven and nobody owns the standard. Two engineers use AI heavily with great results, three use it occasionally, one produces fast code nobody trusts. Without a shared review standard, the fast engineers create risk and the careful ones create friction.
- The uneasy questions go unanswered. Leaders worry out loud about hollowing out junior talent and about losing control of their own systems. Those worries are legitimate, and a memo does not address them. A working practice does: what gets delegated to AI, what always gets human review, how juniors learn in an AI-assisted workflow.
What the teams that get faster do differently
Where the mandate lands, someone owns three documents. The list of task types the team delegates to AI. The review standard that AI-produced code has to pass before it ships. The path a developer follows from occasional use to daily practice.
These documents are boring, none of them takes more than an afternoon to draft, and they are what stands between an adoption percentage and a delivery improvement. They also answer the safeguard the exchange CEO named himself: humans in the loop, with the review standard written down instead of assumed.
The fastest way to get there is to embed it, not to announce it
Standards written in a document change little. Standards demonstrated in the team's own codebase, on real tickets, by a senior engineer working alongside the team, change behavior in weeks. That is the reasoning behind the forward deployed engineer model: a senior engineer embedded in your team who ships features like anyone else, introduces AI tooling in the daily workflow, builds the review playbook with your team, and transfers the practice to your developers before rolling off.
The memo settles that the team will use AI. What it means in practice gets decided ticket by ticket, in code review, over the following weeks, by whoever is in the room.
Frequently asked questions
Should I mandate AI adoption in my engineering team?
A mandate settles the question of whether the team adopts AI, and that clarity has value. The mandate works when it comes with an owned practice: explicit delegation rules, a review standard for AI-produced code, and someone accountable for both.
How do I measure whether AI is actually making my developers faster?
Measure delivery, not usage: cycle time from ticket to merged code, review turnaround, and rework rate, compared before and during adoption. Percent adoption tells you the tools are installed; only the delivery numbers tell you they work.
What is the biggest risk of an AI mandate?
Uneven usage with no review standard: fast, unreviewed AI code entering the codebase while careful engineers slow down to compensate. The fix is a shared playbook for what gets delegated and how output is reviewed, owned by a senior engineer.
How does a forward deployed engineer help with an AI mandate?
A forward deployed engineer turns the mandate into practice from inside the team: shipping in the codebase, configuring the tooling, building the review playbook, and raising the whole team's AI fluency on real tickets until the practice stands on its own.
Part of: Forward Deployed Engineers at Viaro