A playbook pack for AI agent teams

Ship on Autopilot:
Stop prompting AI manually

Bots plan, build, and review each other's work. You say "yes" only at the gates that matter: merge, deploy, release, and every message that leaves your team.

See what is inside

The pack is not available yet.

  • 11 modules
  • 9 role cards
  • 7 human gates
  • 1 zip of markdown files
One change, from plan to ship
  1. PlanPlanning agents set the job.
  2. BuildBuild agents write small pull requests with tests.
  3. ReviewReview agents check the work.
  4. GateThe bot stops. You say "yes".
  5. ShipMerge, deploy, or release.

Proof with every result: a passing test, a preview link, or a screenshot.

[Madson to fill: real result. One true outcome from your own setup, with proof.]

What changes

One agent with every job is a risk

Without clear roles, one AI agent can get every job. Then nobody owns the result, and nothing stops a bad change.

Without clear roles

  • One AI agent gets every job.
  • Nobody owns the result.
  • Nothing stops a bad change.

With this pack

  • Each bot has one job, one lane, and hard stops.
  • Build agents write the code in small pull requests with tests.
  • Review agents check the work. A gate post lists what is still open.
  • Every result comes with proof: a passing test, a preview link, or a screenshot.
  • Bots stop and wait at seven gates. You approve merge to main, production deploy, and store release.
  • Bots draft messages to people outside the team. You push Send.
  • Jobs that work once become scheduled jobs. They stay quiet when nothing changed.

Every rule comes from a setup that the author runs every day.

[Madson to fill: real result. A before and after from your own work, with proof.]

How it works

Run an AI agent team that plans, builds, reviews, and ships your code

The bots check each other's work before you see it.

  1. The roles

    Nine bot roles. Each role has one job, one lane, and hard stops.

  2. Bots that review each other

    Build agents write the code. Review agents check it. A gate post tells you what is still open.

  3. Proof before "done"

    Every result comes with a passing test, a preview link, or a screenshot.

  4. The gates

    Seven points where a bot must stop and wait for a human "yes".

A human only at key gates

You do not watch every step.
You approve the steps that can hurt you.

Merge to main

No change goes into main without your "yes".

Production deploy

No deploy goes to production without your "yes".

Store release

No build goes to a store without your "yes".

Outside messages

Bots draft messages to people outside the team. You push Send.

The pack has seven gates. The full gate table is inside.

Fit check

Is this pack for you?

It is for you if you are

  • A software engineer who wants bots to do routine engineering work.
  • A tech lead who wants bot work with clear owners and human gates.
  • The founder of a small team who wants one person to run many bot lanes.

It is not for you if

  • You want bots with no human gates.
  • You have no code project.
  • You do not want to pay for an AI agent platform.

Inside the pack

11 modules, in the order you set them up

Module 0

The Map

Know what each part does and the order to set it up.

Module 1

One Bot, One Job

Give every agent a job spec with a fixed shape and hard stops.

Module 2

The Roster and the Gates

Nine roles, clear lanes, and seven points where only you can say yes.

Module 3

The Right Brain for Each Job

Match build work and thinking work to the right class of AI model.

Module 4

Code Work Stays in Isolated Build Agents

Get small pull requests with tests and proof, not direct edits on your machine.

Module 5

Visible Handoffs, No Mixed Context

Every handoff happens in the open, and a rule keeps each project's context apart.

Module 6

Bots That Review Each Other

Review agents check the work, and every result arrives with proof.

Module 7

One Quality Bar for Every Bot

Simple code, small changes, tests, and proof before "done".

Module 8

From One Good Run to a Scheduled Job

Turn a job that works once into a job that runs on a schedule.

Module 9

The Release Lane

One bot ships the builds and keeps an exact record of what is out.

Module 10

Plain Bot Writing and a Clean Workspace

Bot replies in simple English, and a rule to clean up secrets after each job.

What you unlock after purchase

  1. The full stack list: the exact tools behind this setup, and the plan level that each tool needs.
  2. The setup steps for each tool in the stack.
  3. The model split: which class of model does the build work, and which does the planning and review work. Exact settings included.
  4. Nine copy-paste role cards for your AI agent team.
  5. One shared rules block that every agent obeys.
  6. The gate table and a merge gate post template.
  7. The quality bar that every build agent must meet, and how the author sets it up.
  8. Scheduled job examples: a weekly status deck and a build watch.
  9. A release lane with a release record template.

Format

  • One zip of markdown files.
  • One intro file, one how-to file, and one file for each module.
  • Copy-paste blocks between START COPYING and STOP COPYING.

Before you buy

What you need

Read this list first. The pack does not work without these items.

  1. A paid subscription to a commercial AI agent platform. The pack names it inside. Its plans cost extra. They are not in the price of this pack.
  2. One code project with a repo.
  3. Coding skill. You must read a pull request and approve or reject it.
  4. A source doc for your project, for example a requirements file.

The setup steps are written for one specific AI agent platform. You can adapt the rules to other tools. The setup steps may not work as written on other tools.

[Madson to fill: typical monthly cost of the required paid plans, from the tool makers' public price pages.]

Who made this

Madson Cardoso

Madson is a senior software engineer and fractional CTO in Lisbon, Portugal. He has built iPhone apps since 2008, when the App Store opened. He has more than 10 years of experience with web and mobile apps. He shipped consumer apps, SaaS companion apps, and B2B tools for teams in Europe and the US. His recent work includes a yoga app and an inventory SaaS for restaurants and groceries. He trained more than 100 developers to use AI in their workflow. He still writes and ships code every week.

Every rule comes from a setup that the author runs every day.

FAQ

Questions before you buy

Do the bots work with no human?

No. The bots stop at the gates. You approve merge to main, production deploy, store release, and outside messages.

Is it for people who do not code?

No. You must read and approve pull requests.

Which tools does it use?

You get the full stack list inside. It uses commercial AI agent tools that need their own paid plans.

Do I pay for anything else?

Yes. The AI agent tools need their own paid plans. You also need a code repo.

Can I use the AI tool that I have now?

Maybe. The rules and role cards work as text in most AI agent tools. The setup steps are for one specific platform.

What results can I expect?

This page makes no result claims.

[Madson to fill: real result, with proof and permission]
Is there a team license?
[Madson to fill: license tiers. Proposed: Solo and Team.]
What is the refund policy?
[Madson to fill: refund policy. Check the EU withdrawal rules for digital content.]
Who made this?

Madson Cardoso, a senior software engineer and fractional CTO in Lisbon, Portugal. Read the full bio in the author section.

Ship on Autopilot

Bots plan, build, and review. You approve at the gates.

  • 11 modules
  • 9 role cards
  • 7 human gates
  • 1 shared rules block

The pack is not available yet.