You Built the App. Who's Shipping It?

Ray Garcia

Published April 16, 2026

You prompted your way to a working app. Maybe you used Cursor, Claude Code, or Copilot. The app works. Now what?

Somebody has to manage the issues. Open the PRs. Handle code review. Merge to main. Deploy. And when a reviewer says "fix this" — somebody has to go back, revise, and push again.

That's the engineering process underneath your app. And if you're building with AI, that process is probably the part you didn't sign up for.

The gap between "it works" and "it ships"

Here's what happens after your AI writes the code:

1. The code needs to land in a PR
2. Someone reviews it — catches bugs, suggests changes
3. You fix what the reviewer found and push again
4. It gets approved and merged
5. It gets deployed to production

Most of us stop at step zero. The code exists in a file somewhere. Getting it through the pipeline? That's a different skill set entirely.

What if the pipeline ran itself?

That's what the slashbin-ai-foreman (opens in a new tab) does. It's an open-source daemon that watches your repos for approved work and ships it — autonomously.

Here's the cycle it runs, continuously, every few minutes:

graph LR
    A["Reconcile"] --> B["Revise"]
    B --> C["Implement"]
    C --> D["Branch Sync"]
    D --> E["Promote"]
    E --> A

Reconcile — finds code that's been committed but has no PR, and creates one.

Revise — picks up reviewer feedback on open PRs and fixes it. This always runs before new work — because finishing what's in flight matters more than starting something new.

Implement — grabs approved issues and writes the code using Claude Code. Up to 3 issues per cycle.

Branch Sync — keeps your branches aligned after code ships to production.

Promote — creates PRs to push finished work from development to production.

This runs across multiple repos, in the background, while you sleep. You wake up, your backlog is smaller.

You don't need to understand all of this

That's the point. The Foreman reads the issue, reads the repo's context file, and invokes the right skill for that repo. You define what to build. The Foreman handles how to ship it.

How a full AI engineering team works

The Foreman isn't a solo act. It's one layer in an AI engineering team. Here's how the whole thing fits together:

graph TD
    PO["Product Owner"] -->|"creates issues"| EM["Engineering Manager"]
    EM -->|"approves with label"| F["Foreman"]
    F -->|"invokes skills"| S1["Service Repo 1"]
    F -->|"invokes skills"| S2["Service Repo 2"]
    F -->|"invokes skills"| S3["Service Repo 3"]
    S1 -->|"opens PR"| R["Reviewers"]
    S2 -->|"opens PR"| R
    S3 -->|"opens PR"| R
    R -->|"feedback"| F
    F -->|"promotes to"| PROD["Production"]

Each layer has one job:

• The Product Owner decides what matters and writes the cards
• The Engineering Manager breaks epics into tasks, validates they're ready, and labels them approved
• The Foreman picks up approved work, invokes Claude Code on each service repo, and opens PRs
• Reviewers — human or AI — check the work and leave feedback
• The Foreman revises based on that feedback, then promotes approved code to production

This is the exact team that builds www.slashbin.io (opens in a new tab). The Product Owner, Engineering Manager, and Foreman are all AI agents. The service repos each have their own implementation skills. Humans set direction and review — everything in between is autonomous.

Built for the real world

A few things that matter when this runs in production:

• Failure cooldowns — if a repo fails twice in a row, the Foreman skips it for a few cycles before retrying. No infinite loops.
• Dual-token security — the Foreman can't approve its own PRs. It uses one token to create PRs and a separate one for the Engineering Manager to approve them.
• Graceful shutdown — if you stop the Foreman mid-work, it waits for in-progress implementations to finish before shutting down.
• Discord notifications — optional. Get status updates in a channel. The Foreman works fine without it.

Get started

git clone https://github.com/xrgarcia/slashbin-ai-foreman.git
cd slashbin-ai-foreman
npm install && npm run build
npm start

Point it at your repos, label your issues approved, and let it work. It uses your Claude Code CLI subscription — no per-run API costs.

The Foreman is open source and MIT licensed (opens in a new tab).

Your AI wrote the app. Now let an AI ship it.

www.slashbin.io (opens in a new tab) — webhook ETL, built by an AI engineering team.