01
Claude Code
Works with me
Takes whatever is urgent. Plans it, builds in its own branch, gets an independent review, ships it.
AI coding agents for small dev teams
Your repo,agent-ready.
I run my own products with a team of AI agents. In one week I install the same system in your team’s repo: your conventions, a review agent on every pull request, and real tickets merged by Friday.
- Alex“Not very cohesive. One idea throughout. Full redesign if you see a better way.”
- Claude CodeNew worktree claude/cohesive-redesign from main
- Claude CodeRewrote the page around one idea. lint, typecheck, build: pass
- Review agentReviewed the full diff: 8 findings, one being this log predicting its own release
- Claude CodeFindings fixed. The build now writes the last rows of this log itself
- GitCommit 88f153a built for production
- VercelThis build is the page you are reading
Real log, Taipei time. The last rows are written by the build itself. After the sprint, the same loop runs in your repo.
01 / The system
What I run every day, and what the sprint installs.
One person.
Three kinds of agents.
02
Codex chats
Admin in parallel
Accounts, dashboards, forms, settings. The work that eats an afternoon runs beside me instead.
03
Server agent
Always on
Builds a written spec overnight and leaves results in the morning, not questions.
Every change, the same loop
- 01Ticket
A real task from the backlog, written so an agent can finish it.
- 02Worktree
Every change starts in its own branch. Main is never touched directly.
- 03Build
The agent follows the repo's rules: conventions, tests, what not to do.
- 04Review
A second agent reviews the full diff. Findings get fixed before merge.
- 05Ship
Merged, released on purpose, checked live. Each step leaves a record.
The rules live in plain files the agents read every time: how we branch, review and ship. That is what makes it reliable, not the model.
Notes: how I set up a build-and-check loop with GPT-6 Astra02 / Proof
Products built and shipped this way, with one person behind them.
Built to be used.
SurgrAn AI-agent-native social growth loopCreate with your agent, review every destination, publish on your terms, and learn from what actually performed.LiveNichePeekFind business ideas hidden in RedditUse real conversations from Reddit to spot problems and ideas worth exploring before you spend weeks building the wrong thing.BetaMemexYour universal content librarySync books, video, and audio into one thinking tool built natively for iPhone, iPad, and Mac.BetaHealth BridgeConnect health signals to work patternsConnect Apple Health data with your work patterns so you can see which habits help you feel better and do better work.Beta
03 / The sprint
Fixed price · One week · You keep everything
Installed in
one week.
I install Claude Code or Codex in your repo so the agents follow your conventions, a review agent checks every pull request, and real tickets from your backlog ship during the week. It is the same setup that builds and ships my own products, with one person behind them.
Agent setup sprint
$1,500fixed price, five working days
- A baseline from your GitHub data, so the before and after is real.
- Your conventions written into the agents' instructions, with two of your engineers.
- Three skills for the weekly work that takes your team 30 minutes or more.
- Guardrails: permissions, hooks and branch protection, so an agent can't do damage.
- A review agent on every pull request, calibrated to your codebase.
- A cost plan with caps on seats and tokens.
- Three real tickets from your backlog merged during the week.
- A recorded handover, short videos, 14 days of support and a re-measure at day 30.
Not included: training a whole team, product features, CI beyond the review agent, and other repos. Each of those is a separate conversation.
How it works
- A 20-minute call
We look at your repo and how your team works today, and pick the three tickets.
- Fixed scope, in writing
Scope, price and what's out of scope before anything starts. You own everything built.
- One week in your repo
Conventions, skills, guardrails and the review agent go in as pull requests you merge. The three tickets ship alongside.
- Handover and support
A recorded walkthrough, then 14 days of support and a re-measure at day 30.
04 / Questions
What teams ask before they say yes.
- We could set this up ourselves.
- You could; the docs are free. Most teams don't, because they're at capacity. A week with someone who runs this every day gets you past the stall.
- What access do you need?
- The least that works: a branch and pull requests you merge, on your own AI seats. Access is removed when the sprint ends.
- What about seat and token costs?
- A cost plan with caps is one of the deliverables, not an afterthought.
- How do we know the gain is real?
- A before and after from your own GitHub data. No 10x promises.
05 / About
Alex Morris
A US founder building iOS, macOS and web products, with AI agents doing most of the typing. I set the direction, write the rules and review what ships.
I got here by running the setup on my own work first: every change in its own branch, a second agent reviewing the diff, releases on purpose with a record. The sprint is that setup, installed in your repo, by the person who uses it every day.