The recurring 500 in your orders list has been in the backlog since August. Everyone knows about it. Nobody has an afternoon.
Engineering Fleet is six agents that pick that work up: find what is breaking, file it, write the fix, test it on staging, and stop exactly where your merge rules say stop. We're taking 10 startups as design partners, free for 3 months.
We built it for ourselves first. It runs on Lua's own platform and it works on Lua's own code.
What the six do
Each agent has one job and hands its work to the next.
The Reliability Engineer reads your error logs, groups recurring failures, traces each one to the code and the recent commits, and files a ticket with a root cause.
The Ticket Master vets what comes in, closes duplicates, ranks by severity and hands the next one to the builder.
The Ticket Pilot picks up the ticket, works in its own sandbox, opens a pull request and follows it through CI and review.
The Staging Sentinel deploys that pull request to staging and exercises it before and after the change: happy paths, edge cases, and the things that should not have moved. A regression goes back to the builder. A pass is recorded against the exact commit it tested.
Then merging follows your branch protection. Checks, reviews, signatures. How much happens without a person is a setting you turn up.
The Security Engineer
This is the one most teams buy twice a year and wait weeks for.
The Security Engineer runs black box penetration testing to SOC 2 standard across all of your edges, continuously, not once a quarter in a window you booked in March. It probes the endpoints the way an outsider would, with no map of your internals.
It also turns dependency and code-scanning alerts into tickets on the same board as everything else, with the advisory, the affected path and the version that fixes it.
So a finding does not sit in a PDF waiting for someone to triage it. It becomes a ticket, the Ticket Master ranks it, the Ticket Pilot opens the pull request, and the Staging Sentinel proves the fix before it reaches your branch protection rules.
The sixth one is the interesting one
Agents that only do tasks stop when something around them breaks.
The Engineering Manager's job is the team. It reads each agent's status and the board, so it can see which tickets are stuck and where the flow is backing up. When three runs stall on the same error, it works out whether the cause is in the tickets or in the builder agent itself.
If it's the agent, it writes a code change to that agent, stages it as a new version and opens a pull request. You release it, or you don't. And a run that dies, or a ticket nobody started, goes back on the backlog with a note saying why.
You decide how far it goes
New agents start in dry run: they report what they would do, without writing to your board or your repositories. You authorize each connection yourself and choose which repositories are in scope. There are caps on tickets per run and per day, and a stop switch on every agent.
The rules live in code, not in a prompt.
Applying
You need a product in production, a maintenance backlog worth the name, and someone who can grant repository access. Roughly 5 to 200 engineers, any industry.
GitHub, Linear and Better Stack logs work today. If you're on Jira, GitLab or something else, apply anyway and tell us: those answers decide which ones we build next.
We set it up on your stack with you, you use it on real work, and every couple of weeks you tell us what it got right, what it got wrong and what you want next. Pricing isn't set. We'll share it with you before the three months are up, and you decide then.