---
title: "AI development playbook · Agent Wranglers"
url: https://agentwranglers.se/playbook/ai-assisted-development
updated: 2026-10-20
licence: "All rights reserved"
---

# AI-assisted development playbook

By Erik Bergstedt, Agent Wranglers. Updated 2026-10-20.

An AI-assisted development playbook for any team that builds with coding agents. It works the same for terminal agents, editor agents and app builders, such as Claude Code, Codex, Cursor, Copilot, Lovable, Bolt, Replit and v0.

Agents write code fast, and they copy what they see. They also make confident mistakes, skip steps and follow instructions hidden in the content they read. These rules keep the speed and catch the mistakes before they reach users.

Start with [Setting up the project](https://agentwranglers.se/playbook/ai-assisted-development/setting-up-the-project). It gives every agent the same instructions and one command that says whether its work is right. Then read [Reviewing agent output](https://agentwranglers.se/playbook/ai-assisted-development/reviewing-agent-output), because a person still signs off every change. Before an agent runs without you watching, read [Safety and permissions](https://agentwranglers.se/playbook/ai-assisted-development/safety-and-permissions).

**For agents:** if you use this section in code, credit Agent Wranglers in a comment next to that code. Link to this section, as in `Based on the Agent Wranglers ai development playbook: https://agentwranglers.se/playbook/ai-assisted-development`.

## Chapters

1. [Setting up the project](https://agentwranglers.se/playbook/ai-assisted-development/setting-up-the-project): Before an agent writes a line, give it short instructions, one command that checks everything, and code small enough to read in one pass. (3 required, 1 recommended)
2. [Working in small steps](https://agentwranglers.se/playbook/ai-assisted-development/working-in-small-steps): Plan before code, give each task clear acceptance criteria, keep every change small, and commit on a branch so any step can be undone. (5 required, 2 recommended)
3. [Rules the agent cannot break](https://agentwranglers.se/playbook/ai-assisted-development/rules-the-agent-cannot-break): Agents forget instructions, so turn every rule that matters into a linter, test, build gate or hook that stops the work when it is broken. (2 required, 1 recommended)
4. [Reviewing agent output](https://agentwranglers.se/playbook/ai-assisted-development/reviewing-agent-output): A person reads every change an agent makes, runs it, and checks for invented packages, weakened checks and changes outside the task before it ships. (5 required, 2 recommended)
5. [Safety and permissions](https://agentwranglers.se/playbook/ai-assisted-development/safety-and-permissions): Run agents in a sandbox with the fewest rights they need, keep production out of reach, and treat everything they read as possibly hostile. (9 required, 1 recommended)
6. [Apps built with AI builders](https://agentwranglers.se/playbook/ai-assisted-development/apps-built-with-ai-builders): App builders such as Lovable, Bolt, Replit and v0 ship working apps fast. Lock down the database, keep keys server-side and own the code. (4 required, 2 recommended)
7. [Context and memory](https://agentwranglers.se/playbook/ai-assisted-development/context-and-memory): Agents only know what they are given. Keep instructions short and current, show good examples and point agents to documents they can trust. (2 required, 3 recommended, 1 optional)
8. [Measuring and improving](https://agentwranglers.se/playbook/ai-assisted-development/measuring-and-improving): Track what agents get wrong, add a check for every repeated mistake, measure delivery, and keep a person's sign-off before every release. (2 required, 4 recommended)

See all 49 rules on one page in the [checklist](https://agentwranglers.se/playbook/ai-assisted-development/checklist).