Crypto LLM retweetledi
I am the VP of People who wrote our AI-First Hiring Policy. The one that requires every manager to prove AI cannot do a job before they're allowed to request headcount.
I got the idea from Shopify's CEO. He posted a memo saying teams need to demonstrate why AI can't handle the work before asking for more people. Our CEO forwarded it to me with one word: "This." I had a policy drafted in 48 hours. Approved in one meeting. No pilot. No committee. Seven pages, twenty-three questions, effective immediately.
The form asks managers to document a 90-day AI automation pilot before any requisition is reviewed. Attach transcripts. Provide measurable outcomes. Explain what you tried, when, and why it failed.
Nobody completes it. That's the point.
Since January, 340 headcount requests have come in. I've approved 11. The rest get a templated response: "We encourage you to explore AI-augmented workflows before resubmitting. Resources available in #ai-transformation." That Slack channel has nine posts. Six are mine. One is a vendor who somehow got added.
A director in customer success actually tried. She ran Claude on escalation calls for two months and submitted her documentation. The AI told a customer disputing a $4,200 charge to "consider whether this frustration might be an invitation to reexamine your relationship with material attachment." She sent me the transcript with the subject line "Here is my proof."
I denied the requisition. Insufficient pilot duration. Policy says 90 days. She'd done 60. I told her the early results showed "promising engagement with the tooling" and encouraged her to continue.
She quit in March. I listed her departure under "natural efficiency optimization."
I personally tried using AI once. I asked it to draft a standard headcount denial. It wrote: "We understand the strain your team is experiencing, and want to assure you this decision was not made lightly." I deleted it immediately. Way too empathetic. I replaced it with what we actually send: "Please reference the AI-First Framework (Section 4.2) and resubmit when assessment criteria are met."
The AI had more compassion than my actual process. I noted that as a limitation of the tool.
Here is what's actually happening. The average employee at our company now covers 2.3 roles. I know this because I calculated it for a board slide. I presented it as "AI-enabled capacity expansion." The board loved that phrase. It sounds like technology. It's actually just people.
Nobody says "I'm doing three jobs." They say "I'm still figuring out how to leverage AI for this." Because saying you can't handle the workload means admitting you haven't tried the tools. Admitting you haven't tried the tools means questioning the policy. Questioning the policy means a meeting with me.
One director did question it. At an all-hands. In front of 200 people. He asked if anyone in leadership had personally used AI to replace a function of their own role.
I enrolled him in a Growth Mindset Workshop. It's 90 minutes. It goes on your file. He resigned three weeks later. That's one more headcount we didn't have to deny. It self-resolved.
My own team is fully staffed. People Operations is exempt from the AI-First policy. "Human nuance required for sensitive organizational matters." I wrote that clause. I also approved three new hires on my team last quarter — an AI Ethics Coordinator, a Workforce Transformation Analyst, and a Change Adoption Specialist. Their job is to help everyone else do more with less.
Meanwhile, we posted a job listing last week. "Head of AI Strategy." $280,000 base. Reports to me. We received 1,400 applications in 48 hours.
We are simultaneously running a hiring freeze and hiring.
The board just approved my promotion. Chief People & AI Transformation Officer. $90,000 raise. They cited my work in "avoiding $18 million in headcount costs." I calculated that number by taking every denied requisition and multiplying it by average fully-loaded comp. That's not savings. That's arithmetic performed on a thing that didn't happen. But on a slide it looks like savings. On an investor call it sounds like strategy.
I still don't know what AI can actually do. I spent eleven minutes with it and it was nicer than me.
But I know what the policy is for. People do the work. The headcount goes down. When headcount goes down after you say "AI," the AI gets credit. No one checks who's actually doing it. No one will. Because checking means admitting the policy doesn't work. And the policy is working beautifully.
The people doing three jobs are the proof.
If they weren't here, we'd need to hire. Since we didn't hire, something must be handling it. As long as nobody looks too closely at what that something is, it's AI.
English
