Operator memo: Git tracks the change. The question is no longer whether you can ship. It is how you cut the damage.
Building is no longer the bottleneck
Companies already have the history: the PR, the commit, the author, the merge, the deploy. An engineer can open it and reconstruct what was touched, and that is enough for the code. It is not enough for the business that comes next.
AI already writes to production. Sometimes a human is prompting. Sometimes an agent has permission to ship on its own. Sometimes it is the internal team, a vendor, and that agent in the same week. Buying another phase, another build, another project does not fix what breaks after that: whether what went out is okay, and how you cut the damage when the change already hit production. The software still matters. Writing it is no longer the scarce work.
The scarce work is operating that path: who looks, with what, before it stays live.
GitHub stores the diff, Vercel stores the release, Linear stores the task. Each one does its job, and none of them is the operator. The statement is there. Treasury is not.
What breaks
A request comes in over chat. Someone prompts, or an agent with production access takes it. There is a PR. Someone merges. The deploy goes out. GitHub has the trail of the change, Slack has the request, Vercel has the fact that it is live.
Two weeks later there is an incident, and the question in that room is not only who said yes. It is whether what went out was okay: what was being asked for, whether it came from a human prompting or from an agent allowed to ship, whether QA, security, legal, or a human looked at what that execution did not see, or whether nobody decided that a look was needed.
You can reconstruct it, which is why it looks solved, but it is scattered: nobody was operating over the whole path. There was history, and there was no operator.
That is where quality goes, and continuity too. Another build can already be generated in-house from a prompt or from an agent. What another phase does not give you is a way to cut the damage when AI is already writing to production.
Ask, operate, look
Asking, operating, and looking were always there. Sometimes the same person asks and looks. What expanded is operating: AI already writes, because someone prompted or because you gave an agent permission to ship to production.
What cannot stay implicit is the look. It is mandatory, or you ask for it, or you decide this change does not need it. That look can be a QA agent, a security agent, a legal agent, a human, or more than one of those. An agent can ask. An agent can operate. The same run should not silently be all three. If there is a look, it has a name: not the PR approve, but whoever (a person or a specialist agent) sees what the prompt or the autonomous run did not see.
Production is not the output of the chat. It is a system someone has to keep operating after the PR is already merged, with less damage, not just more shipping.
How we are thinking about this
At WakeUp Labs we build and operate systems in production. Companies show up prompting, or with agents that can ship. AI writes. Sometimes they ship with nobody holding the path, and nobody asked if what shipped was okay. The gap is not that a tracker was missing. The gap is that there is no orchestration when the work is already moving that way.
That is why Operator exists: not to replace GitHub, Vercel, or Linear, and not to invent a history that already exists. It is the orchestration of the operation. The request, how it ran (a human prompting, or an agent with permission to ship), the look (QA, security, legal, a human, or none, on purpose), production, and what comes next. That is what a company needs when AI already writes to production and the question is how to cut the damage.
That is Operator. It works, and we are implementing it with clients. It lives here: operator.wup.ar.
It is not self-serve. We are opening a whitelist so you can try it before everyone else. If this is your problem, write to contact@wakeuplabs.io with the subject Operator, tell us which system you operate, and we will consider it.
The three questions
In under a minute, with evidence, not from memory:
Was the last change that reached production okay, and what was being asked for?
Did it come from a human prompting, or from an agent with permission to ship?
Who looked (QA, security, legal, a human), or did you decide nobody needed to?
Anyone can ship from a prompt or from an agent. Nobody should do it blind, and nobody should confuse a PR with having an operator.