Please Approve the Staffing Plan

I had been using /plan mode to research a new evaluation system. Then I worked with Codex to define a /goal for actually executing the project.

For weeks, I had been experimenting with loops and agent orchestration. This time, we got very clear about what “done” looked like, and I sent Codex on its way until it reached the first point that needed my validation.

A couple of hours later, I checked in.

Six of nine steps complete. Great.

The next step read:

Obtain approval from Cris for staffing plan and budget.

I did a double take.

I’m sorry. Staffing plan? Budget?

I steered the task and asked Codex to explain exactly what it wanted me to approve. The answer: permission to “employ” 12 agents across 21 role types, with an unlimited token budget and a mix of models and reasoning-effort levels.

It had not encountered a technical blocker. It had apparently reached headcount planning.

This was a first for me. I am used to reviewing and approving staffing plans for humans. I had never approved one for a workforce of AI agents preparing to execute my project.

Twelve agents. Twenty-one role types. Unlimited tokens.

Apparently, my next project artifact is an org chart.

What a time to be alive.

---

This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.