Journal
Here I write about how I work with coding agents and the questions that come up along the way. When my approach changes, I update the note.
Introducing Claude Code to a development team: rules, review, costs
What I would settle before a team uses coding agents day to day, and what I know about it from my own setup.
Running MCP servers safely: permissions, secrets, prompt injection
How I limit my own and third-party MCP servers in my agent setup, and where those limits end.
Keeping the token costs of coding agents under control
How I combine subscriptions, daily budgets and routing so agents do not get expensive without anyone noticing.
What I learn from operating AI agents
Why I treat routing, handoffs and reviews as parts of the same system.
How I write tasks for agents
A task needs context and a result that can be checked.
What I put in agent instructions
Project-specific rules need a source and a clear scope.
What context I give agents
Relevant sources, open questions and handoffs between sessions.
How my agent platform, Harness, routes tasks
Routing by cost, remaining quota and failure history.
The boundaries I set for AI agents
Limit access and enforce rules in the system.
How I assess AI tools
A concrete task is a better starting point than a feature list.