Ex-Google engineer built a graph-based workflow that made Claude Code do nearly 10 days of work in about an hour. Here's the exact setup he uses. step 1. create a separate Git worktree for every task. run: git worktree add ../task-1 -b task-1 open a new terminal, cd into... it, and launch Claude Code. repeat this for every agent. each one gets its own branch, so nothing conflicts. step 2. give every Claude agent a different task. same repository. different branches. they all work in parallel. step 3. once they're done, review the pull requests. keep the good ones. merge what works. step 4. add a review agent. paste each PR into Claude and ask: "does this implementation match the original specification" every agent now has another agent reviewing its work before it gets merged. step 5. scale it up. 48GB of RAM can comfortably run around 20 Claude Code agents. 128GB can handle roughly 50. most engineers run Claude Code once and wait. he runs dozens of Claude Code agents in parallel, then merges the best results. this 60-minute lecture is one of the best free resources on graph engineering and multi-agent workflows. bookmark it and watch it below.show more

At 20 agents, launching them isn't the hard part. Knowing which one needs you before merge time is.

This is exactly how you build in codex by default. Its support for worktrees is amazing

Bookmarkedddd will watch when I’m free

Helpful

I’m happy to hear that

damn great step by step guide

worktree 这招真挺适合并行折腾的

This is a great example of using AI to parallelize software development instead of treating it like a single coding assistant. Breaking work into independent task and adding a review step makes the workflow mouch more scalable and reliable.

oh wow

worktrees plus a review agent is the part people skip when they brag about agent headcount

All I see is merge hell.

This is the kind of workflow optimization that makes me think we're still in the early days of understanding how to best

Good orchestration beats throwing everything at one agent.

cutting 10 days of work down to an hour is insanely effective

stack bloat feels productive lol tbh

“每个任务一个 git worktree” 这一步特别关键，因为后面不只是并行，更是验收不串。一个 worktree 对一张任务单、一个 diff、一个 rollback 点，review 时连 comment 漂移都会少很多。很多人觉得 graph engineering 难，先把这层隔离做好，后面的 agent 分工才不会越跑越乱。

the review agent is my favorite part

i do this with a distributed control plane that coordinates all the work onto many machines-scales horizontal and vertical and can use pi with local or openrouter or nim free models even and/or claude code-u can mix n match and it autoscales and autoheals

Oh man this is a huge Graph Engineering Masterclass

Graph-based workflow = 10 days in 1 hour. the prompt engineering grind is over. the prompt orchestration monopoly is starting.

let’s get it

Alot of steps.

using separate git worktrees for each task is clever. keeps the claude agent isolated so one failure doesn’t derail the other nine.

smooth workflow here mate

What they built is interesting

insane work jesus.

wow insane bro bookmarked