Essays & field notes

Writing

Agents, systems, experiments, and the knowledge that makes intelligence useful.

07Featured / Agent Systems

Most Agents Are Bullsh*t

Most agent projects break in the work around the model: defining the job, constraining its actions, evaluating failures, and improving it from production.

Read essay

August 8, 2026 / 6 min read

All field notes

07 essays

06EvaluationEvals Are Bullsh*tMost evals measure whether an agent looks good. Good evals measure whether it does the job your business actually needs.August 1, 20266 min read

05Agent SystemsThe Model Is the Wrong Thing to OwnEnterprises should own their intelligence. That means accumulating domain knowledge every new model can inherit, not training a model of their own.July 24, 20265 min read

04EvaluationSalesBench: The Long-Horizon Agent-to-Agent EvalA long-horizon RL environment where a small model learns to manage an insurance sales pipeline against an LLM buyer, scored by revenue closed instead of by an LLM judge. The trained model vastly outperforms the untrained base, and the gap widens as the eval gets harder.May 14, 202612 min read

03Agent SystemsThe Agent Research LoopWhat Karpathy's autoresearch really means, where agent systems are headed, and an open-source harness that ran 550 experiments over a weekend.March 16, 20268 min read

02Personal AgentsI Run a Personal AI Agent 24/7 on a Mac Mini. Here's How It Actually Works.A Mac Mini, some markdown files, and seven communication channels. Inside the setup that gives me a 24/7 AI assistant that monitors my email, iMessage, WhatsApp, and Twitter - and actually does useful things.March 7, 202612 min read

01Model TrainingI Let AI Agents Train Their Own Models. Here's What Actually Happened.Two frontier agents, a pile of bugs, and a reality check on the future of autonomous AI research.February 8, 20267 min read