Blog

Notes from the data layer

Short pieces on the ideas behind the projects: memory, money, time and what agents do with them.

1. 9 Mar 2026 · 5 minThe AI agent illusionAgents forget the brief, won't replan, and once looped for eleven days on a $47,000 bill. The model isn't the broken part. The system around it is.
2. 21 Jan 2026 · 4 minHow LocusGraph remembersAn agent that fixes the same bug twice never kept anything. LocusGraph stores what it learns as linked memories, filters what gets in, lets old ones fade and recalls by walking the graph.
3. 8 Dec 2023 · 3 minTerminal UX: why command lines are making a comebackApps get louder every year, yet developers spend more time in the terminal. It shows what happened, keeps a record of what you did, and lets small tools snap together.
4. 12 Sept 2023 · 3 minIntent, not instructions: what blockchain systems should understandOn a blockchain you translate every wish into exact instructions. Intent-based systems take the goal instead, and the hard part is reading it right.
5. 24 Jun 2023 · 4 minCode is thought written downCode has two readers: a machine that needs every step and a person who needs to know why. Writing for both is the job.