The "AI-First" Memo Won't Save You

Somewhere right now, an executive is drafting a memo that says AI usage is no longer optional. Reflexive AI usage will be a factor in performance reviews. Headcount requests must first justify why AI can't do the work. I've read a half-dozen versions of this memo. I understand the impulse behind every one of them.

And I've yet to see the memo, by itself, move the needle on anything except the metric it names.

Mandates produce compliance, not adoption

Here's what happens when you mandate AI usage and start measuring it: usage goes up. That part works every time. Engineers open the tool, run a prompt or two, and let the telemetry record a daily active user. If Copilot suggestions-accepted becomes a review metric, accepted suggestions climb — including the ones people accept and then quietly rewrite. You asked for usage; you got usage. Goodhart's law doesn't take a day off just because the technology is exciting.

What a mandate cannot produce is the thing you actually wanted: engineers restructuring how they work because the new way is better. That change is pulled, not pushed. It happens when a developer watches a teammate hand a well-scoped ticket to an agent, review the diff over coffee, and merge before standup — and thinks, I want that. No memo has ever manufactured that moment. Plenty of memos have poisoned the well before it could happen, because the tool now arrives smelling like surveillance.

The irony is that demand for AI usually already exists — it's just routed around you. Shadow AI numbers make this hard to ignore: one BlackFog study found 86% of employees using AI for work tasks weekly, with 58% admitting to unapproved tools. Your people aren't resisting AI. They're resisting your AI — the sanctioned path with the clunky login, the wrong model, and the acceptable-use doc nobody can parse.

What executives should mandate instead

I'm not against executive force. Some things deserve a mandate — they're just not the behavior itself. They're the conditions that make the behavior rational.

Mandate the paved road gets funded. Someone has to own tool selection, configuration, shared context files, and internal patterns as a job, not a hobby. If your "AI-first" push has no budget line and no named owner, it's a vibe, not a strategy. The mandate worth issuing: every team gets a working, pre-configured, security-approved setup within 30 days, and there's a person whose OKRs suffer if that's not true.

Mandate that friction gets reported and killed. The honest question isn't "why aren't you using AI?" It's "what happened the last time you tried?" Usually there's a concrete answer: the agent can't reach the internal package registry, the repo has no context file so output is garbage, legal took three weeks to approve a tool that was obsolete in two. Executives are uniquely positioned to destroy this kind of friction. That's the highest-leverage use of the corner office in this whole transition.

Mandate outcome measurement, not activity measurement. Decide up front what better looks like — cycle time on well-scoped work, review turnaround, defect escape rate — and instrument it before the rollout, not after. If leadership reviews outcomes, teams optimize the work. If leadership reviews usage dashboards, teams optimize the dashboard. Pick which economy you want.

Mandate that skeptics get a real forum. Your most valuable adoption signal is a senior engineer who tried the tool seriously and can articulate exactly where it fell down. Under a usage mandate, that person learns to keep quiet, and you lose the ground truth you need to fix the rollout. When engineers are pushed to maximize AI throughput without guardrails, workflows get built fast, then stall at compliance review — and leadership concludes AI "doesn't work here." The memo caused the failure it was meant to prevent.

The tell

There's a simple test for whether an AI-first push is real or theater: ask what changed besides the expectation. New budget? Named enablement owner? Reworked security review SLA? Different metrics in the QBR deck? If the only artifact is the memo and a usage dashboard, you've built a compliance program, and you'll get what compliance programs get — minimum viable participation, gamed numbers, and a quiet consensus among your best people that leadership doesn't understand the work.

Adoption at the org level looks exactly like adoption at the individual level: someone tries the tool on a real task, it works, and they never go back. Your job isn't to command that experience into existence. It's to make it nearly impossible to avoid.

Takeaway: Don't mandate the behavior; mandate the conditions. Fund a paved road with a named owner, hunt friction like it's your job (it is), measure outcomes instead of usage, and protect the skeptics who tell you the truth. The memo that matters isn't "you must use AI." It's "we've made it worth your while — here's what changed."