The agent reached production: eleven words from one night
On-call chat, the small hours of Thursday
On-call 03:12. Prod is throwing 500s, orders have lost the status column. Last commit is the agent's, an hour ago.
Tech Lead Who let it into prod at night?
Backend Me. Migrations from CI, headless mode, YOLO mode. The prod key was in the env.
On-call Doing a rollback. Checkpoint is at 02:00, we lose forty minutes of orders.
Tech Lead Roll it back. And turn self-healing off, it's gone in twice to "fix" the schema.
Backend Off. I'll rebuild the forty minutes from the logs by hand.
On-call And who would a permission prompt have gone to at night? Me? I was asleep.
Tech Lead Call at ten. The key comes out of CI before then. Migrations go through a sandbox, and not at night.
On-call Kill "accept all" while you're at it. Nobody opened the migration diff: tests green, ship it.
Tech Lead That's not the agent. That's us.
On-call Rolled back. Going to bed.
At three in the morning nobody asks what a word means: rollback first, sleep second. The words stay in the chat till morning.
At night the argument is about the rollback. In the morning it's about access
Half of these words are about how the agent ended up in production on its own. The other half are about what gets put between it and production the next morning.
Headless modeModo headless
Running an agent with no interactive UI, from a script or command line, for automation.
In headless mode the agent ran nightly in CI with no one watching.
How an agent gets run where nobody is watching it. Hence "who let it in".
YOLO modeModo YOLO
A setting that lets an agent make edits and run commands without asking you to approve each step.
He ran the agent in YOLO mode and it refactored, tested, and committed on its own.
The name gives the attitude away: that's what the people who switch it on call it, and the people who clean up in the morning.
Permission promptSolicitud de permiso
An agent's request for your go-ahead before an action like writing a file or running a command.
A permission prompt appeared before it deleted the folder, and she declined.
The window YOLO mode does away with. At night there's nobody to answer it — that's where the incident began.
RollbackReversión
Undoing an agent's changes to return the project to its earlier state.
The refactor broke the build, so she triggered a rollback to last night's state.
The first thing you do at three in the morning. The arguing comes later.
CheckpointPunto de control
A saved snapshot of project state you can return to if an AI's later changes go wrong.
Before the big refactor I set a checkpoint to roll back to.
Where to roll back to was decided the evening before, when it was set. Forty minutes of orders is the distance from it to the incident.
Self-healingAutorreparación
An agent's ability to detect and fix a failure, like a broken test, on its own.
The self-healing pipeline caught the crash and patched it before anyone woke up.
Reads as a virtue on a tool's feature list. At three in the morning it gets switched off: it keeps fixing the thing you just rolled back.
SandboxEntorno aislado
An isolated environment where AI-generated code or agent actions can run without touching the real system.
The agent tested its script in a sandbox, so the risky delete never reached real files.
Where migrations get sent after one of them made it to production.
Human-on-the-loopHumano sobre el bucle
A design where a person oversees an agent and steps in on problems, rather than approving each action.
The team stayed human-on-the-loop, watching the dashboard and intervening only when it drifted.
A role nobody in the chat named, and the on-call engineer played it all night. Differs from human-in-the-loop by how many times a night you get woken up.
Accept allAceptar todo
Applying every change an AI proposed at once, without reviewing them one diff at a time.
The suggestions looked fine, so she hit accept all and moved on.
The button you press when there are more diffs than patience. It's also where people look when they trace where the incident began.
Ship itA producción
A rallying phrase for releasing quickly, often after AI did the heavy lifting and the author trusts it enough to deploy.
Tests pass, looks fine — ship it.
A rallying cry, not a decision. In the morning it gets quoted in a different tone of voice.
Automation biasSesgo de automatización
The tendency to trust an automated or AI output without checking it.
Automation bias let a wrong number sail through because 'the AI said so'.
"That's not the agent, that's us" — this is it, and the chat never called it by name. A word about the people around the agent, not the agent.
At ten in the morning these words become a plan
At night they were what people said to each other. On the call they'll be what gets written down: the key, the sandbox, who watches the overnight runs.
Questions and answers
What about the rest of the words in this field?
On the deck page, all of them. These eleven are the ones that came up in one night.
Why is there Spanish on the card?
In the chat it's "rollback" and "sandbox"; in a Spanish team's postmortem it's "reversión" and "entorno aislado", and a search for one won't find the other. Both forms sit on one card.
Where do the definitions come from?
From the Vibe Coding & AI Agents deck — the same cards as in the app.
The next incident will bring a different eleven.
The Vibe Coding & AI Agents deck holds the rest — in the app they come a few a day.
