The Director of AI trap

A pattern we see almost weekly: a company decides to “get serious about AI”, opens a single senior role — Head of AI, Director of AI, VP of AI Engineering — and hires for it. Six months later, the same company calls us. The system is half-built, the lead is overwhelmed or already gone, and the board is asking pointed questions about ROI on a hire that has not yet produced ROI.

The Director of AI role, treated as a starting move, has a structural problem. Here is the version we wish someone had told us before we watched this happen the first three times.

One person, six jobs

Production AI is six disciplines at once: applied ML engineering, prompt and eval engineering, data engineering, security and compliance review, product management, and change management with the team whose work the AI will affect. The unicorn who is genuinely strong in all six does not exist. The candidates closest to that profile already have jobs they are not leaving.

So the role gets compromised on hire. You get a great ML engineer who is not a strong product manager, or a great product person who has never owned an on-call rotation. The thing they are weakest at becomes the bottleneck on every project they touch. And because they are a director, nobody above them in the org has the technical depth to see it happening until the timeline has slipped twice.

Experience is dosage, not credentials

A senior AI hire’s CV looks impressive. Past employer, past project, conference talk, published paper. None of that tells you what we actually need to know: how many production AI systems have they personally been on-call for? How many times have they watched an eval set save a release? How many times have they walked into a room of skeptical operations staff and convinced them the agent is a tool, not a layoff plan?

For most senior individual hires, the honest answer is: two to five times. For a team that has built AI for clients across industries for the last few years, the answer is more like fifty. Pattern recognition compounds with reps. A single hire, no matter how brilliant, is at the bottom of that curve at the moment you most need to be near the top.

The political cost nobody budgets for

A new Director of AI walks into an org chart on day one. They need to negotiate for engineering time, data access, security review, marketing buy-in, and budget against a CFO who has not yet seen an AI invoice come back below estimate. The first three months are political work, not engineering work, and the political work is far harder for a new hire than for a partner team that arrives explicitly framed as “helping the existing leadership ship.”

We see this play out in calendar invites. The internal Director of AI has back-to-back stakeholder meetings for the first quarter. They are not shipping. They are introducing themselves. By the time they are clear of that, the budget review has happened.

The handover argument, properly framed

There is a real argument for an internal AI lead: long-term ownership, institutional memory, the ability to integrate AI decisions with the rest of your product roadmap. Those arguments are real. They are also long-horizon arguments — they matter in year two, not week six.

The version of this story that works: hire the AI lead into a working system. Let a partner stand up the readiness audit, the anchor build, the eval suite, the runbooks, the documented architecture, and the on-call rotation. Then write a Director of AI brief that is “extend and own this,” not “build this from nothing alongside your other twelve jobs.” That brief attracts a different — better — candidate, at a lower salary, with a faster ramp.

The cost difference between those two paths, in our experience, is roughly half a million pounds and nine months. It is the most consequential strategy decision most companies will make about AI this year, and it is almost always made unconsciously, in a board meeting, as a recruiter brief.