I'm Saad — I build systems for a living. This one's free, because it's useful.
AGENTS
Build a self-improving AI system with Claude
Here's the whole thing — the 14 steps that turn "I want an AI that does the work" into a real agent that runs itself, reaches for the right tools, handles its own failures, and hands you something you can actually ship.
What's inside
- The thing nobody tells you
- What it looks like when it works
- Step 1 — Define the agent (write its system prompt)
- Step 2 — Pick the model
- Step 3 — Give it skills + tools
- Step 4 — Set permissions (least privilege)
- Step 5 — Configure the environment (network allowlist)
- Step 6 — Preinstall the packages it needs
- Step 7 — Create a session
- Step 8 — Load it with resources (files, repos, data)
- Step 9 — Wire credentials through a vault
- Step 10 — Kick off the task
- Step 11 — Handle events (watch its actions; steer or stop it)
- Step 12 — Turn on observability (session tracing)
- Step 13 — Let it self-recover (read failures, patch, continue)
- Step 14 — Review the diff/output before you ship
- The bigger picture
- Your next step
Open the guide PDF · free · no email needed
save it, share it, use it
save it, share it, use it
If you run a business — I build these
For agenciesThe Agency Command CenterEvery client, one screen — health scores, dying-ad alerts, one-click reports.For sales floorsThe Sales Command CenterEvery call logged, revenue traced to the setter, an AI analyst on top.For DTC brandsThe Content & Landing-Page FactoryLanding pages and ad scripts generated — you approve, it ships.