What if I joined your AI team tomorrow?

Days 1–30 · Listen and map

No reorganisations, no tool mandates, no strategy deck in week one.

Find where AI work already happens. In most organisations it already does — informally, unevenly, sometimes invisibly. I meet the domain teams, the sceptics included, and map what exists: experiments, shadow tools, stalled pilots, and the people quietly doing good work without a mandate.
Understand what "working" would mean. For each candidate use case, I ask the domain owner one question: what outcome, verified how? If nobody can answer, that use case is not ready — and knowing that early is cheap.
Set the measurement baseline. Before anything ships, agree how we will judge it: task completion, review and rework cost, defect escape, safe execution, end-to-end time, and cost per accepted outcome. Usage is a signal; outcome is the result.

Deliverable at day 30: an honest map — what exists, what is ready, what is theatre — and one chosen lighthouse.

Days 31–60 · Prove one path

Ship one lighthouse use case end-to-end, with the team that will own it — not a demo owned by me. The domain team keeps the meaning: workflow, acceptance criteria, final sign-off. I bring the delivery rhythm, the technical path, and the guardrails.
Put the hard rules in the system. Permissions, approval gates for irreversible actions, audit trails, cost visibility. The prompt guides; code owns the hard stops. This is the discipline I practise on my own infrastructure, applied at company scale.
Start the shared road, minimally. Only what the lighthouse actually needed: an evaluation habit, a controls pattern, cost tracking. Shared platforms earn their existence by being extracted from real work, not designed in advance.

Deliverable at day 60: one use case in real use, with evidence, owned by its team — and the first reusable patterns written down.

Days 61–90 · Make it repeatable

Second and third teams adopt the road. Coaching over gatekeeping: I transfer the patterns so capability grows beyond one builder — the same handoff discipline I applied when moving my own PoC work to delivery teams.
Establish the operating rhythm. A management view of quality-adjusted productivity: useful delivery after review, rework, defects, safety and cost. Leaders should never have to choose between believing the demo and believing the numbers.
Decide what not to scale. Some pilots should end at 90 days. Saying so, with evidence, is how the portfolio stays credible — and how the organisation learns that "no" from the AI team is information, not obstruction.

Deliverable at day 90: a repeatable path from idea to adopted system, at least one thing deliberately stopped, and a scoreboard leadership trusts.

What I will not do

Promise headcount replacement to win sponsorship.
Ship ungoverned pilots to show speed.
Pull ownership away from domain teams into a central bottleneck.
Chase the model of the week while the evaluation question goes unanswered.

Measure me on this

At day 90, ask four questions: Is one AI use case in genuine production use, with its team owning it? Can we see its quality and cost honestly? Is there a written, reusable path for the next team? And did anything get stopped for the right reasons? If the answer to all four is yes, the foundation is real — and the second 90 days get much faster.

The plan flexes to what I find; the principles do not.