Stop planning your code. Ditch MCPs. Commit to main. Says the creator of OpenClaw — and I can’t totally disagree.
It’s a pretty (seemingly) radical approach:
- No spec documents
- No plan mode
- No branching
- No issue trackers
- Commit directly to main
- Spec-driven development was a crutch for uncertainty
- He doesn’t even read most of the code his agents write
Heresy!!! Most of those statements could start flame wars on their own. But hear him out:
He’s shipping faster than entire engineering teams.
What Changed His Mind?
Plan mode was a crutch for weaker models. Today he just starts a conversation with the AI, explores the problem together, and when it feels right, types one word:
Build
MCP was a crutch for tool integration 👀. It bloats context with data the model doesn’t need and can’t be chained or scripted. CLIs let agents filter, compose, and verify — closing the loop naturally in bash.
He builds iteratively. Plays with it, feels it, sees how it behaves, then refines.
He was out of the game for three years. Burned out. Early retirement. Hadn’t touched a terminal. Came back with fresh eyes and built what many believe Siri should have been.
Albeit what he built is incredibly dangerous without guardrails, but it has changed the direction of the world. Changed the stock prices for companies. 178K stars… one of the fastest growing products in the history of mankind.
This Isn’t How We Were Taught
We were taught to plan exhaustively, spec thoroughly, branch carefully, and review line by line.
But what if the best developers of the next era look nothing like that? What if they’re more like product thinkers who feel their way through problems at inference speed?
Trusting the model the way a director trusts a cinematographer? The tools have changed. Maybe the process has to change with them.
Most engineers aren’t behind on skills. They’re behind on unlearning.
What’s the one thing in your workflow you know you should let go of but haven’t yet? Drop it below.
