Hire the leader who has shipped production AI.

You have AI work running in several places and no single view of what it cost or what it returned.

Every AI effort inventoried and ranked, what each one costs to run, where the measurement is missing, and a stop list. Each one tested against three conditions: can the source be traced, can the exception be owned, can the savings be booked.

You can name your bets, what they cost, the ones you stopped, and which of the three conditions to fix first.

Your roadmap comes from the loudest inputs, and nobody on the team has spoken to a customer this quarter.

A weekly customer touchpoint the team runs themselves, the signals worth acting on separated from the ones that are noise, and the assumptions behind the next bet tested before it is committed.

A standing customer conversation your team runs without you, and the next bet chosen out of it.

AI writes the first draft of most things now, and nobody on the team can say whether the output is good enough to ship.

Your quality bar pulled out of your own approval history into a scored test set, one workflow measured against it, and human review kept on the paths where a wrong answer costs money.

A test set your reviewers run on real work, and a decision they changed because of it.

Your agent demos well and has never run in front of a customer.

What production needs and a demo does not. Data the agent can read, a permission model scoped per customer and per role, and the measurement that says whether it worked.

One agent live on a real workflow, the gate that keeps it there, and the cost per run.

Your agents answer the same question differently depending on which system they read.

Ingestion across the systems you already run, the same customer resolved to one customer everywhere, classification and metadata so the data can be cut by something other than a keyword, and retrieval on top. Rights and limits enforced in code.

One set of data your product and your agents both read, with the rules enforced in code.

Your team ships more than it ever has, and the roadmap moves at the speed it did last year.

Your teams classify their own work. What moves to agents on a schedule, and what keeps a named owner. The rhythm and the ownership come from your team, and the design serves it.

One team running a weekly cycle you can watch, and the next team named.

You are writing next year’s hiring plan and the roles no longer mean what they meant when you last wrote one.

Roles, levels and ratios re-cut against what agents now produce, the bar rewritten for judgment, and an interview loop that tests for it.

An org chart you can defend, a hiring plan naming the roles you stopped opening, and a loop that has run.

Your AI costs rise with usage and your pricing does not, so every new customer costs you margin.

Cost per unit of work the AI produces, inference tiered by what the task is worth, packaging and pricing rebuilt on those numbers, and the margin modelled as usage grows. Spend carried as a share of payroll, which is the one denominator a finance team can budget against.

A price that holds margin as volume rises, and AI spend stated as a percentage of payroll rather than a number with no base.

Your most senior people are the constraint and the work queues behind them. Engineering, environmental, professional services, field work.

One workflow first. The expertise the work depends on becomes an agent that drafts, with sign-off kept human.

An agent drafting real work in production, and the next workflow chosen.