At a glance
WHAT IT’S REALLY ABOUT
How Codex Goals enable long-running autonomous work with verification loops
- Codex Goals differ from normal prompts by running an autonomous loop of execute–verify–decide-next-step until evidence shows the outcome is achieved.
- Effective Goals are outcome-based and measurable, pairing a clear finish line with verification methods, constraints/guardrails, scoped boundaries, and an iteration/stop policy.
- Claire shares a real five-hour-45-minute autonomous coding run and shows how /goal can systematically burn down recurring production issues instead of applying one-off fixes.
- Technical demos focus on using logs (Sentry/Vercel) to categorize errors, identify root causes, ship fixes via PRs, and re-validate against historical events until errors reach zero.
- Non-technical demos show /goal as an operations assistant—cleaning thousands of emails and mass-triaging Linear tasks—highlighting that long-running goals can improve everyday workflow hygiene.
IDEAS WORTH REMEMBERING
5 ideasUse /goal when you’re stuck in “keep going / try next / rerun tests” mode.
Goals shine when you’d otherwise micromanage an agent step-by-step; the loop self-directs and keeps iterating until it can prove completion.
Write goals as measurable outcomes, not vague tasks or refactors.
“Refactor this code” or “make customers happy” lacks a finish line; better goals specify what should be true and how success will be validated (benchmarks, tests, log replays, etc.).
Strong goals combine verification with guardrails to prevent “cheating.”
Claire emphasizes constraints like keeping correctness tests green—otherwise an agent could hit the metric by breaking or removing functionality (e.g., deleting a slow page).
Include boundaries and an iteration policy so the agent knows where to look and how to proceed.
Specify allowed tools/files/systems (Sentry/Vercel/checkout subsystem) and require iteration reporting (what changed, what evidence showed, what experiment is next).
Treat production error logs as a backlog you can systematically eliminate.
Her Sentry example uses /goal to categorize root causes, implement fixes, and replay historical events until “invalid operation” errors go to zero—yielding a more coherent framework, not scattered band-aids.
WORDS WORTH SAVING
5 quotesIf you find yourself in that process, using /goal in Codex might be a tool that you wanna add to your toolkit.
— Claire Vo
With Goal, when you give Codex a goal, it actually has something that it can work towards, and it will continue to loop to the next step and verify until it can measure that it has met that goal.
— Claire Vo
But the first time I used Goal, I was actually able to get a coding task running for about five hours and 45 minutes, which is longer than I've ever had anything run before.
— Claire Vo
You really wanna use Goals when you would otherwise find yourself saying the same thing after turn," like, "Keep going," "Try the next thing," "Run it again," "Now run the test," "Continue until it's actually done."
— Claire Vo
Goals are strongest when it has three properties: a durable objective, an evidence-based finish line, and a path that may require several turns of investigation.
— Claire Vo
High quality AI-generated summary created from speaker-labeled transcript.
