Pinned post
This is f*cking insane. Anthropic pays the engineer who built Claude Code $1M+ a year, and he hasn't typed a single line of code since the day he uninstalled his IDE
In this 95-minute interview, he walks through the exact setup that ships 30+ pull requests a day
Prompts → Plan Mode → Parallel Agents → Self-Review Loop → Code You Never Type
step 1 → Run 5 Claudes at once - 5 tabs, each on its own copy of the repo, so the agents never step on each other
step 2 → Start agents from your phone - he kicks off a few the moment he wakes up, and they're already coding in the cloud before he opens a laptop
step 3 → Plan mode first, every time - go back and forth until the plan is right, then it one-shots the build almost every time
step 4 → Let it test itself - it runs the tests, writes new ones and launches a copy of itself to check it still works
step 5 → Claude reviews every PR - it catches about 80% of bugs before a human looks, and every mistake that repeats gets turned into a lint rule
Claude Code now writes 90% of the code at Anthropic, and 100% of his
One month, 10 to 20 PRs a day, not one line edited by hand. The agent shipped 2 bugs. By hand it would have been 20
Most engineers still type every line themselves
You don't have to
Bookmark it and watch it tonight
Then read how to wrap a loop like this into a full harness in the harness engineering guide below ↓
In this 95-minute interview, he walks through the exact setup that ships 30+ pull requests a day
Prompts → Plan Mode → Parallel Agents → Self-Review Loop → Code You Never Type
step 1 → Run 5 Claudes at once - 5 tabs, each on its own copy of the repo, so the agents never step on each other
step 2 → Start agents from your phone - he kicks off a few the moment he wakes up, and they're already coding in the cloud before he opens a laptop
step 3 → Plan mode first, every time - go back and forth until the plan is right, then it one-shots the build almost every time
step 4 → Let it test itself - it runs the tests, writes new ones and launches a copy of itself to check it still works
step 5 → Claude reviews every PR - it catches about 80% of bugs before a human looks, and every mistake that repeats gets turned into a lint rule
Claude Code now writes 90% of the code at Anthropic, and 100% of his
One month, 10 to 20 PRs a day, not one line edited by hand. The agent shipped 2 bugs. By hand it would have been 20
Most engineers still type every line themselves
You don't have to
Bookmark it and watch it tonight
Then read how to wrap a loop like this into a full harness in the harness engineering guide below ↓
95:43
Your agent passed every testView quoted post on X
The issue is still open
A better prompt won't fix it, because the model isn't the broken part
I broke down the harness, loop and graph around it: what each one owns, how they nest, and which one to fix first ↓ x.com/i/article/2118…
