Featured dispatchJul 7, 2026 7 min read
Why We Break Agents on Purpose
We found that the only way to build truly robust autonomous systems is to subject them to chaotic, adversarial environments.
Read the dispatch02 / Field notes
Lab notes.
Technical notes on operating AI agents, evaluating failures, and building reliable context systems.
Archive
02 entries02Jun 22, 202603May 14, 2026
The Architecture of the Crumbles Agent Layer
How Tracey and Atlas form the platform underneath Clara and Riley—the full-stack products that put agents to work.
What Production Multi-Agent Systems Actually Need
Multi-agent systems are becoming distributed software. The hard problems are no longer prompts—they are contracts, context, observability, and controlled autonomy.