Re-prompt instead of hand-editing
Technique
When an agent's code is wrong, change the prompt or context and rerun instead of fixing the output yourself. The idea is that some input produces code needing no edits, and once you find it you can repeat the task without new effort.
Theo's current stance
Theo likes this idea, quoted from Alex in Lauren's essay. He compares it to a speedrunning trick: try many odd approaches, and once one works you never have to think about that task again.
Try it
- When your agent's output is wrong, change the prompt or context and rerun instead of editing the code by hand.
- Once a prompt works, save it so the same task can be repeated without new effort.
First seen
19:11
In theory, there's a series of input tokens that's going to give you the code that you want as an output that doesn't need editing. It's entirely possible. You just need to figure out that series of input tokens.
Every mention
19:11
In theory, there's a series of input tokens that's going to give you the code that you want as an output that doesn't need editing. It's entirely possible. You just need to figure out that series of input tokens.
19:24
I like this framing a lot. It's almost like a speedrunning technique. Like you could just go through the part the normal way or you can try a shitload of weird stuff and if you get it right once then you never have to think about…
Theo Daily is unofficial and not affiliated with Theo or t3.gg. Written by Claude Opus 5.5 from each video's transcript.