Skip to main content

No prior agent experience assumed

Start with one delegation. End ready to trust it.

This is a hands-on path for engineers who have not used Droid, or any coding agent, in a codebase where mistakes are expensive. You change one thing at a time, see what it does, and learn why each control exists before you need it.

Start here

A terminal, Node 20, and a Factory account. No AI background needed.

Learn by breaking

Every control you add, you then try to get around. That is how you find out what it is worth.

Finish with evidence

A scoreboard that shows what each change bought, which is the thing a risk reviewer will ask for.

The only mental model you need first

You describe an outcome. Droid decides the steps, and the harness decides what it may do.

Droid is an agent. You give it a task in plain language. It reads the repository, picks a tool (read a file, edit one, run a command), looks at the result and picks again, until the task is done or it needs you. Everything it did is recorded in the session.

larkspur-payments
droid "Document POST /transfers in docs/api.md"

The harness is everything around the model: what it reads before it starts, what it may do without asking, and what it cannot do at all. Most of this workshop is about the harness, because that is the part you control.

session
One task, start to finish: every prompt, tool call and result. Resume it, fork it, audit it.
tool call
One action: Read, Edit, Create, Execute a command, or an MCP tool. The real record of what happened.
autonomy
Off, Low, Medium or High: what runs without asking you first. Ctrl+L to change it.
harness
AGENTS.md, settings, rules, hooks, skills, droids and MCP. Files in your repo, versioned and reviewable.