Ship While You Sleep - 5 Claude Code Agents in 30 Days

Build the Overnight Fleet with Agents That Resume After a Crash, Never Collide and Hand You Finished Work by Morning.

De: J Cook

Narrado por: Virtual Voice

You left five agents running overnight. One failed silently. One retried 47 times and burned $50. You shipped nothing. Again.

Meanwhile, developers on Reddit are running fifty agents against one database at the same time and waking up to finished work. The difference isn't talent. It's a ledger their agents write to before they act.

In 30 days you will turn one babysat agent into a fleet of five that ships while you sleep. This weekend you get the first win: kill a running agent at step 4 of 6, restart it, and watch it finish at step 5 without redoing a thing. By day 7 two agents work the same repo all night and never touch the same file. By day 14 five agents pull from one queue and clear a backlog you would have worked through alone. By day 21 the fleet sends planning to the expensive model and execution to cheap ones, so the same budget buys you more agents. By day 30 you close the laptop and read the trace in the morning.

Twenty eight books on this shelf teach you to configure one agent inside one session. Not one of them survives process death. The strongest of them costs $22.39 from a major publisher and stores your agent's memory in markdown files. Its own buyer says only about a third of it was useful. None of them ship a schema. None of them get you a second agent working safely. This is the book those authors needed before they wrote theirs.

Every agent, table and drill in this book comes with a free companion GitHub repo: the finished fleet package, a runnable drill for all twelve chapters, and one command that kills a live agent and resumes it. No key, no account, no install.

Here's what you'll build:

1. A five agent fleet pulling from one shared queue, with heartbeats and a dead letter queue you can inspect.
2. An agent.db with three contracts: control state, an append only event ledger, and derived knowledge with provenance.
3. A resumable session wrapper that takes a hard kill mid task and finishes the job on restart.
4. An idempotent execution layer with receipts, so the third retry never repeats the side effect.
5. A claim registry with leases, so two agents work one repo without overwriting each other.
6. A cost governor and routing split that buys you more parallel agents out of the same monthly spend.
7. A replay driven regression suite seeded from your own production failures, plus a 30 day operating plan.

This book is 180 plus pages of runnable machinery. Scroll up and grab your copy.