Signals from Noise

We are only months into the era of reasoning-capable models being used for real engineering work.

The space moves fast, here's where I write things down as I figure them out.

Series

demystifying llms

A series exploring the mental models necessary to build reliable systems with Large Language Models. We strip away the chatbot metaphor and reconstruct LLMs from first principles: tokens, context windows, sampling, and the generation loop.

7 entries view_series >

Logs

attention is a budget

#context #retrieval #architecture

year of the harness

#engineering #orchestration

thought completion engines

#mental-models #foundations

the planning bottleneck

#engineering #ai-proficiency #architecture

software has always been orchestration

#mental-models #architecture #orchestration

the emergence of agent skills

#engineering #orchestration #context

the architecture of truth

#engineering #architecture

tool calls are structured output

#mental-models #engineering #orchestration

sampling, not retrieval

#llms #retrieval

system prompts are just text

#mental-models #engineering

context is everything the model sees

#engineering #context

generation is a loop

#llms #foundations #engineering

the chat illusion

#llms #foundations