writeups

23 posts, newest first

2026

08-29
Your agent is only as good as your repo1min
08-21
temperature is a TypeError now, and the API sends a 400 for it anyway1min
08-20
You probably don't need multi-agent3min
08-14
agents can't read your website, and the fix is one Accept header5min
08-09
The price crash quietly turned your 2024 cost hacks into debt2min
07-24
MCP went stateless and my infrastructure diagram got boring2min
07-17
OpenAI is un-shipping fine-tuning, check your calendar1min
07-12
MCP in production, now that it grew up3min
07-08
Design tools for the model you have, not the intern you imagine4min
07-07
Agent memory that isn't a vector database cope4min
06-28
The small model was the right call and everyone knew it3min
06-25
Track LLM spend like cloud spend or repeat cloud history3min
06-12
What AI code review catches, misses, and hallucinates2min
05-25
Streaming is a UX decision you keep making by accident2min
01-23
Sandbox the agent like it's someone else's code, because it is4min

2025

12-24
Nine engineer portfolios worth stealing from2min
12-10
Prompt caching is the only discount you control2min
11-15
Structured outputs are API contracts, act like it1min
11-14
Your context window needs a budget, not a bigger limit4min
10-31
Write the eval before the prompt, you animal2min
10-20
Fine-tuning is the last resort and that's a compliment4min
08-23
Humans in the loop, but at the right loop3min
08-19
Your RAG problem is a retrieval problem wearing a trench coat3min