The Best Designs Come From One Prompt
If you want the best design an LLM can produce, don't ask for one - ask for five in a single call, and force the model into parts of latent space it would never visit cold.
John Higgins
Writing about technology, AI, software, and product.
Latest posts
Don't Let One Model Grade Its Own Homework
A model's review uses the same weights that produced the code. Swap models between write and review - different priors catch what self-review can't.
Domain Code Reviews
Stop asking LLMs to 'review this code.' Run domain-specific passes - edge cases, concurrency, resilience, security - and catch what generic review misses.
What To Do With Your Extra Tokens
Surplus AI coding tokens are wasted capacity. Use them for targeted code reviews, UI audits, product analysis, and marketing - with copy-paste prompts for each.