Series
AI Product Sense
Judgement calls about building AI products: what to ship, what to charge for, what agents change
This series is one post right now, and I'd rather say that plainly than dress up an empty page. AI Product Sense is where I put judgment calls that sit above the code: what to ship, what to charge for, what an agent actually changes about a product. Calls that don't show up in a benchmark table.
The post here, on DeepSeek's open-sourced agent harness, is the shape the rest will take. The model DeepSeek released is not the interesting part. The program wrapping it, the loop, the tools, the rules about what the agent is allowed to touch, is where the product decisions actually live, and those are the same decisions anyone shipping an LLM feature has to make whether they notice or not. I read the repo package by package instead of taking the announcement at face value.
More lands here as I make more of these calls.
In publication order
Nobody has laid out a reading path for this series yet, so these are listed oldest first — the order they were written in, not a curated one.