Macro 3D Playbook court with an AI agent paused at a human approval gate.
Agent Foundation One useful task · your code · a clear handover

An AI agent for one job your team needs done.

Bring an idea, a repeated task, or a project that stalled. We build and test one useful job. You keep the code and instructions, and we help you make the next change. Going live is a separate project.

  • Scope 1 role · 1 job
  • Ownership Your code and instructions
  • Handoff You make the next change

The handoff

What the first project includes.

We agree on the task, build it, and test the result with you. Your GitHub repository is the folder of code and instructions your team keeps.

Technical handover example

agent-foundation/

Illustrative · not a client result
Illustrative job
Turn meeting notes into a reviewable action list.
Input
Meeting notes
Draft result
Actions with owners and due dates
Approval
A person checks the draft before it is saved to another tool
  • AGENTS.md How Codex should work in the project
  • src/ One useful job, working end to end
  • policy/ Rules for what the agent may do, ask, or stop
  • evals/ One real example and one failure case
  • tests/ Checks you can run again
  • RUNBOOK.md How to start it, its limits, and how to recover
  • CONTINUATION.md The next agreed change

Ownership and production

What you receive, and what a launch adds.

You keep the code, data definitions, instructions, rules, tests, setup guide, known limits, and project history. The agreed test environment needs no hidden CREATE SOMETHING account or undocumented login.

Agent Foundation

One job works in the agreed test environment.

  • The repository includes the source, rules, tests, and runbook.
  • You can see what works, what is blocked, and what needs checking.
  • You complete one change with Codex before handoff.
Production Promotion

Going live is a separate project.

  • Live account access, service connections, and moving data
  • Deployment, rollback, monitoring, and recovery
  • Checking connected tools and testing with real users

We quote Agent Foundation after reviewing the project and agreeing on the job. Launching it is a separate project called Production Promotion. That includes live access, deployment, monitoring, and testing with real users.

Foundation fit

Tell us the job you want the agent to do.

On the call, we’ll agree on the first useful job. If there’s a fit, you leave with the proposed scope and the basis for a quote.

Talk about your project If the role is still unclear, we may recommend mapping the job first.
Project
An idea or an existing project
Role
The person or team the agent serves
Input
One real example of the work
Result
What a good result looks like, including when to stop