Your Company Is Still Level 1 (And the AI-Pilled Memos Aren't Going to Change That)

I have spent the last year walking into companies that say they are AI-pilled, and the pattern is the same every time. Leadership has read the same Tobi Lütke memo. Someone has written the same internal Slack post. The org chart looks identical to the one from 2023. There is a Cursor license layered on top of everything. The agents cannot see the documents. Nobody owns wiring up the MCP server. The one person who has compounding gains is the engineer who quietly built her own thing on the side. Everyone else is using ChatGPT to summarize meetings and calling it transformation.

This is what Ann Miura-Ko of Floodgate calls Level 1, and most companies who think they are far past it are still living in it.

What Does "AI-Pilled" Actually Mean?

On May 1, Miura-Ko published a framework for how organizations actually scale AI inside their walls. Six levels, L0 through L5. Her opening line is the cleanest sentence on this topic I have read all year: "Announcements ≠ adoption."

The L0 picture she paints is uncomfortably specific. A CEO who gives an excellent speech about AI transformation while still running the company through the same executive staff meetings, the same status updates, the same reporting lines, and the same headcount plans. The talk has changed. The operating system has not. That is L0.

L1 is just personal productivity. Saved prompts in someone's notes app. ChatGPT for drafting. The org chart unchanged. Maybe a "Head of AI" hire with a budget but no operating mandate. Her hard test for whether you have actually moved beyond L1: if your best AI user left tomorrow, would their workflow remain in the company? Most companies fail it.

The reason this matters is that the public-market story has compressed the levels. The press-release version of "we use AI" makes it sound like Klarna, Shopify, Block, and Anthropic are all at L4 or L5. The actual day-to-day inside those companies is messier. The day-to-day inside the company reading their memos is messier still.

Why Are the AI-Pilled Memos So Persuasive?

Because they are written to be. They are positioning, not field reports.

Look at the pattern when you line up the headlines next to the corrections.

Klarna told the world in February 2024 that its OpenAI chatbot was doing the work of 700 customer service agents. By May 2025, Sebastian Siemiatkowski was telling Bloomberg, "cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality." The company started hiring contract support staff back. The 700-agent equivalency was the headline. The walk-back, fifteen months later, was not.

Block told employees in March 2025: "None of the above points are trying to hit a specific financial target, replacing folks with AI." Eleven months later it cut roughly 40% of its workforce, and Jack Dorsey told the market that "AI fundamentally changes what it means to build and run a company." Both statements are real. They cannot both be honest descriptions of the same operating posture.

Shopify's Tobi Lütke published the "reflexive AI usage is now a baseline expectation" memo in April 2025, and posted it publicly the same day. The investor audience was the point. Shopify's headcount had quietly dropped from 11,600 to 8,100 over the three years before the memo. Nobody called that a layoff.

Anthropic's CPO Mike Krieger told the room at Cisco's AI Summit in February that "Claude products and Claude Code are being entirely written by Claude," and that internal usage at Anthropic was "effectively 100%." Redwood Research published a skeptical analysis a few weeks later. Their estimate of company-wide AI-merged-line code was closer to 50%, with the higher figures concentrated on the Claude Code team itself. The "100%" line travelled. The methodology pushback did not.

None of these are scandals. They are companies optimizing market perception, exactly as a company should. The mistake is treating the memo as your operating manual. The memo is the brochure. The brochure is not the building.

What Does Real Internal AI Enablement Look Like?

It does not look like installing an MCP server and walking away.

It looks like four things, done in order, with humans in the room.

First, remove the access impediments between your employees' agents and the systems those agents need. Most internal AI friction is not model capability. It is that the agent cannot read the doc, cannot see the dashboard, cannot hit the API. Fix the plumbing first.

Second, sit next to people and solve a real problem together. Daan van Rossum, who studies internal AI champion programs, puts it cleanly: "even when the tools are there, the licenses are paid for, and training has happened, nothing really changes." Adoption is not a tooling problem. It is a people problem. People copy people. The job is to be a workflow translator, not a teacher.

Third, build skills, not assumptions. Anthropic's own framing is the right one to steal: "MCP connects Claude to data; Skills teach Claude what to do with that data." A working MCP without a skill is a phone book without a person to call. The skills get built by working through real problems with real employees, then encoding what worked. Miura-Ko's L4 bar is the target. A sales rep packages her call-analysis pattern as a shareable skill. A CX engineer packages a ticket-investigation pattern. Skills move horizontally across functions, authored by the people who do the work, not by the AI team.

Fourth, train people. Until someone watches an agent do their job for the first time, they do not know what to delegate. You have to work three or four real problems with them before the lights come on. There is no shortcut. The orgs that skip this step end up with what OpenAI's own State of Enterprise AI report calls a "6x productivity gap between power users and average employees using the same tools." Same licenses. Completely different outcomes.

The Bet

By December 31, 2027, at least 70% of Fortune 500 companies will have a public org-chart or careers-page role with "AI Enablement," "AI Champion," "AI Ops," or equivalent in the title. The function is going formal at the upper end of the market. The teams that are quietly winning right now already have someone, full-time, whose job it is to do the four steps in the previous section. The teams still treating AI as a license-purchase decision will be measurably behind by the end of 2027.

If fewer than 70% of Fortune 500 companies post such a role by then, I was wrong. Screenshot this and check the math on January 1, 2028.

You Don't Get Level 5 by Reading About It

Miura-Ko closes her piece by paraphrasing Steve Blank, and the parallel lands. A startup is not a small version of a large company. An AI-pilled company is not an AI-assisted version of an old company. It is an organization rebuilt around a new operating model. Nobody, including her, yet knows what the top of the curve actually looks like.

What we do know is that you cannot get there by reading other people's memos. You get there by going up the levels yourself. Stage by stage. Remove impediments. Sit next to people. Build skills. Train. Watch what works. Encode it. Like a product launch, not a press release.

Open your org chart. If it looks the same as it did two years ago, the memo was the marketing, and you are the audience. The company in the memo has already moved on.

What Does "AI-Pilled" Actually Mean?

Why Are the AI-Pilled Memos So Persuasive?

What Does Real Internal AI Enablement Look Like?

The Bet

You Don't Get Level 5 by Reading About It

Sources