moo.fm Bloghttps://blog.moo.fm/ A journal about building agents you can trust.en Your agent doesn’t need more context, it needs fewer choiceshttps://blog.moo.fm/posts/your-agent-doesnt-need-more-context-it-needs-fewer-choices/https://blog.moo.fm/posts/your-agent-doesnt-need-more-context-it-needs-fewer-choices/I cut one agent’s tool list from nineteen down to six, and it got better at all six. How I decided what to take away, and the one tool I was wrong to remove.AgentsWed, 04 Nov 2026 12:00:00 GMT Getting the same answer twicehttps://blog.moo.fm/posts/getting-the-same-answer-twice/https://blog.moo.fm/posts/getting-the-same-answer-twice/A pipeline that works most of the time is a pipeline you can’t debug. What I pinned down, and what I gave up in order to pin it.DeterminismWed, 21 Oct 2026 12:00:00 GMT What I let an agent touch, and what I don’thttps://blog.moo.fm/posts/what-i-let-an-agent-touch-and-what-i-dont/https://blog.moo.fm/posts/what-i-let-an-agent-touch-and-what-i-dont/The blast radius is the design. Where I drew the line around credentials, customer data, and anything that can send mail.SecurityFri, 02 Oct 2026 12:00:00 GMT If you can’t replay it, you can’t trust ithttps://blog.moo.fm/posts/if-you-cant-replay-it-you-cant-trust-it/https://blog.moo.fm/posts/if-you-cant-replay-it-you-cant-trust-it/I keep every prompt, tool call and result, and I can replay any run. It took a weekend and it has paid for itself twice over.ObservabilityTue, 15 Sep 2026 12:00:00 GMT My agent passed every test and shipped a bughttps://blog.moo.fm/posts/my-agent-passed-every-test-and-shipped-a-bug/https://blog.moo.fm/posts/my-agent-passed-every-test-and-shipped-a-bug/A hundred green evals, and not one of them checked the thing that mattered. What went into the suite afterwards.EvalsThu, 27 Aug 2026 12:00:00 GMT Stop asking the model to be carefulhttps://blog.moo.fm/posts/stop-asking-the-model-to-be-careful/https://blog.moo.fm/posts/stop-asking-the-model-to-be-careful/Adjectives in a prompt do almost nothing. Structure does almost everything. Three rewrites, with the before and after.PracticeThu, 06 Aug 2026 12:00:00 GMT Give the agent a way to say it doesn’t knowhttps://blog.moo.fm/posts/give-the-agent-a-way-to-say-it-doesnt-know/https://blog.moo.fm/posts/give-the-agent-a-way-to-say-it-doesnt-know/Most bad output is a model with no exit. I built one in, and the confident nonsense mostly stopped.AgentsSat, 18 Jul 2026 12:00:00 GMT The job I automated three times before it stuckhttps://blog.moo.fm/posts/the-job-i-automated-three-times-before-it-stuck/https://blog.moo.fm/posts/the-job-i-automated-three-times-before-it-stuck/Twice I automated the wrong half. What finally made it hold, and how I could tell the difference.AutomationTue, 30 Jun 2026 12:00:00 GMT The build log nobody reads is the one you needhttps://blog.moo.fm/posts/the-build-log-nobody-reads-is-the-one-you-need/https://blog.moo.fm/posts/the-build-log-nobody-reads-is-the-one-you-need/I made my agents narrate their decisions. It cost about four percent in speed and saved me weeks.ObservabilityThu, 11 Jun 2026 12:00:00 GMT Where the idea for moo.fm came fromhttps://blog.moo.fm/posts/where-the-idea-for-moo-fm-came-from/https://blog.moo.fm/posts/where-the-idea-for-moo-fm-came-from/It started as a complaint about my own calendar. The version I almost built instead, and why it was wrong.PracticeFri, 22 May 2026 12:00:00 GMT