Topics
Topics
The Lab by theme. Pick a thread and follow it across the streams.
Loop engineering and the agent loop
What the loop actually is, why its stop condition is the hard part, and what an enforced brake looks like in code.
The harness
The runtime around the model, and why the unit of evaluation is harness plus model plus task.
Evaluation and reliability
Why an aggregate score hides what kills you in production, and how to bucket failures instead.
Multi-agent systems
When splitting work across agents helps, when it is theater, and what the measured comparisons show.
Retrieval and context
What long context changed about retrieval, and what it did not.
Skills and tooling
The constraints that actually bind when you wire capability into an agent.
Authority, security, and governance
Blast radius, the lethal trifecta, and authority that is scoped and enforced outside the model.