For the complete documentation index, see llms.txt. This page is also available as Markdown.

Best practices

Start narrow

Do not try to automate everything at once. Start with one repeated motion that matters.

Build for real behavior

Pick an agent your team will actually use, not just admire.

Be specific

Clear triggers, clear context, clear outputs, clear guardrails.

Deliver results where people already are

Slack, email, and inbox artifacts often work better than expecting people to go looking for results elsewhere.

Review outputs before scaling

An agent can save huge amounts of time, but trust comes from testing.

Do not overcomplicate version one

A simple agent with a clear trigger and output is often more valuable than a complex one with too much logic.

The first win matters more than the perfect system.

Last updated