01. creating
8 rules for better docs just shipped—these are all my takeaways on writing docs for AI and with AI in 2025…. And exactly how we’ve scaled docs at Replit.
I also gave a talk at a NextJS Conf side-event on the same topic. You can find the slides here. Thanks to Erik & Andre for a stellar event
02. learning
I get lots of (positive) comments on my audio. Audio Hijack + Loopback have changed how I pre-process. This video by Thomas is an excellent resources for anyone looking to implement
There’s an underrated OpenAI Cookbook on what makes docs good. I have no idea how it took me this long to find
03. thinking
I’ve been seeing barbell strategy in all things lately.
It can be applied to AI coding—you want to avoid “the middle” as much as possible:
Waiting for AI to respond or do research
Stopping and restarting workflows
Struggling with context or prompting
And optimize for the “ends,” that means either
Asynchronous workflows: hands-off prototyping or bug fixes where you rely on AI to handle a large chunk of the work while you take care of more pressing matters.
Flow-state sessions: where you’re tightly integrated with the AI (almost a pseudo-agent mode), editing a small set of files and getting rapid feedback on your applications
As much as possible, you should seek to avoid watching AI and waiting for it to respond and either optimize traditional devex (flow state, feedback loops, & cognitive load) or seek out the novel asynchronous approach.
