Articles
Writing on AI engineering: when to use an agent, how to manage its context, how to orchestrate subagents, and what the vocabulary means.
- Bugs Aren't the Boogeyman AnymoreFor seventy years, the bug was the thing software feared most. That fear is quietly dying, and something else is taking its place.6 min
- How I Got Codex to Orchestrate 50+ Subagent RunsThe setup behind my Luna, Terra, and Sol agents. No custom orchestration framework required.5 min
- Context Shouldn't Be Append-OnlyMost long-running agents have a janitor. What if the model managed that working context itself?7 min
- Stop Calling Everything an Agent FrameworkRuntime vs framework vs harness, using LangGraph, LangChain, and Deep Agents to make the stack click.13 min
- Tokens, Embeddings and Context WindowsAn LLM never sees the sentence you typed. It sees a billable sequence of fragments, a geometry problem, and a position map.Field Notes: AI Engineering, 1.413 min
- The Nine DialsNine decisions define most AI systems before you write a single line of model interaction code.Field Notes: AI Engineering, 1.36 min
- Vocabulary as a Design PrimitiveVocabulary in AI engineering isn’t just semantics. The words drive the decisions.Field Notes: AI Engineering, 1.25 min
- Workflow, Agent, or Multi Agent? Pick OneEveryone talks about AI agents. But most “agents” I see in the wild are a model call with some if-statements wrapped around it.Field Notes: AI Engineering, 1.15 min
- Field Notes: AI EngineeringThe start of Field Notes: AI Engineering, a series on the full stack of AI system design in the order it makes sense to learn.Field Notes: AI Engineering, 0.02 min
- The Builder Who Teaches. The Teacher Who Builds.On Andrej Karpathy and four of his recent projects: LLM Wiki, autoresearch, microgpt and nanochat.4 min