Skip to main content
Back to side projects

Crew Autopilot

Everything on autopilot.
Except what matters.

It carries the work from report to release on its own. And hands control to the right people, only when it matters.

Crew Autopilot: the team around an Agentic Development Lifecycle (ADLC)

01The problem

You are the bottleneck.

Even with a coding assistant.

In an early-stage startup there is often a single developer.

And they don’t just write code: everything in the team goes through them.

You

one developer

  • Userswork out whether a report is really a bug
  • Monitoringfind the cause of a production error
  • Product teamturn an idea into a spec
  • Coding assistanttell it what to do and check the code
  • Testersget the new version tested and collect feedback
  • Teamupdate the docs and answer “where are we at?”
  • Foundership once the go-ahead comes

AI now writes the code. Everything else is still waiting for you.

02The solution

Now there are two of you in the middle.

It does the heavy lifting. Key decisions stay with the team.

Same people, same needs. But you are no longer on your own.

You + Crew Autopilot

you oversee, it executes

  • Usersreproduces the report and works out if it is a bug
  • Monitoringinvestigates the error and prepares the fix
  • Product teamgets the spec ready and approves it
  • Codewrites it, gets it reviewed, passes 7 checks
  • Testersget the version to try and give the outcome
  • Teamfinds docs and status always up to date
  • Foundergets the release ready and gives the go-ahead
the autopilot does ita person decides

03How it works

Follow one ticket through the team.

From request to release, in six steps. Click a step or use the button to move on. Illustrative example.

New ticket

“We’d like users to filter transactions by date range.”

Step 1 of 6

The request arrives

autopilot

Product team

Intake · 09:12

“We’d need a date-range filter on transactions. Can we do it?”

04In the field

Born on a real product.

And built to go further.

Today it develops and maintains Helmo, an AI financial co-pilot in production. Bugs, new features, docs and errors all go through it.

Discover Helmo

Where it is heading

  1. Today

    Develops and maintains Helmo, from request to release.

  2. Next

    Learns from every cycle: problems become fixes to its procedures.

  3. Goal

    A standalone autopilot, able to build products from scratch.

05Let’s talk

It is a work in progress.

“Less Iron Man robots and more Iron Man suits.”

— Andrej Karpathy

It is not a distributable product yet, but I am happy to talk about it: how it works, the choices behind it and what I learned building it.