Agent architecture · Context engineering
How a turn works in an agent harness
Why model calls are stateless, why context and token usage grow, how tools create request loops, and what compaction changes.
Read the explainerWriting and exploration
Ideas at the edge of AI and enterprise systems.
Notes on applied AI, system design, and the techniques that turn emerging research into useful engineering practice.
Featured
Field guides
Evidence-led comparisons and explainers for decisions that look simple on the surface but have materially different architecture, cost, and control implications.
Project Atlas
GraphRAG
Knowledge systems · Architecture landscape
GraphRAG feature and pricing matrix
A practical comparison of eight open-source graph-knowledge approaches, production reference architectures, deployment trade-offs, and the real cost model behind them.
Explore Project Atlas
Agent continuation
Different clocks.
Agentic engineering · Control models
Autopilot, Loop, and Goal
A vendor-neutral field guide to three continuation mechanisms—what makes each agent continue, what makes it stop, and how to choose by the required stopping rule.
Open the field guideArchive
Earlier explorations
Applied AI · Enterprise operations
AI-powered process optimization
An exploration of how intelligent systems can identify bottlenecks, improve decisions, and make operational workflows more adaptive.
Read article
Machine learning · Exploration
Understanding torque clustering in AI
A technical introduction to a physics-inspired approach to identifying structure in complex, multidimensional data.
Read article