xandreed

Series

Reading paths

Ordered, multi-part arcs. Standalone posts live on the home page.

Building a coding agent 9 parts
  1. An agent loop is a while loop with good manners draft
  2. Your app calls a model in more places than you think draft
  3. Your LLM provider is state, not architecture
  4. Subscription OAuth is the hard half of multi-provider auth draft
  5. The model never reads your handler
  6. Most approval prompts should never render
  7. The whole permission model is one folder draft
  8. Error messages are prompt engineering now draft
  9. Your agent is a protocol, not an app draft
Effect in practice 3 parts
  1. Where exceptions are allowed to exist
  2. Make the compiler tell your ids apart: a guide to branded types
  3. The model will find the fourth shape: property-based testing with fast-check
Effect, from zero 3 parts
  1. Effect from zero: the beginner's guide I wanted
  2. Effect, in full: services, layers, errors, and the concurrency toolkit
  3. AI applications are Effect-shaped
How an agent remembers 4 parts
  1. The transcript is the memory draft
  2. The context window is a resource you engineer, not a buffer you fill draft
  3. Sub-agent transcripts are capital, not exhaust
  4. Zero-config storage: Postgres is an option, not a prerequisite draft
Observing an agent 3 parts
  1. Observability is built in; where it goes is a layer draft
  2. Name spans for humans, query them by attribute draft
  3. Your eval has no database; the trace is the dataset draft