Use parallel agents to build a database engine

They split a SQLite implementation across several coding agents working in separate branches, then merged the pieces after repeated test-and-fix loops.

Done with Custom agent· Advanced

What they did
They started with one Claude run to set up the documentation, a Rust crate scaffold, and a test harness. After that, six workers — two Claude, two Codex, and two Gemini instances — operated in separate workspaces, each grabbing a task from a lock file, coding it, running checks against sqlite3, updating shared notes, and pushing changes. A separate Gemini pass was meant to merge overlapping work, but it only happened at the end.
How it went
The project ended up with about 19,000 lines of Rust across 154 commits, plus 282 AI-written unit tests that passed. It also passed 64 SQLLogicTest queries. Support was still incomplete: subqueries and CASE were missing, concurrency was absent, and some rough edges remained; 54.5% of commits were for lock-and-claim coordination, and Codex used an entire Pro weekly quota.
Tips
Parallel agents work best when each one has a narrow, separate job and a shared mechanism for claiming tasks. Expect a lot of coordination overhead, and pause before any merge, push, or other irreversible step so you can review what changed.

Try it yourself with Claude Code

Help me build a [project] in [language] using several coding agents at the same time. First, create a short spec that breaks the work into independent modules such as [module 1], [module 2], and [module 3]. Then launch one agent per module in separate worktrees, and have each agent: pull the latest main branch, claim a task with a lock file, implement the module, run the relevant tests, compare behavior against [reference tool or oracle], and write brief notes on what changed. Before any merge, push, delete, or submit action, stop and ask me to approve it. After the modules are combined, report what passed, what failed, and any coordination problems or duplicated work you found.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on kiankyars.github.io ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.