Essays & Field Notes
Architecture decisions, failure modes, and operational data from running autonomous AI systems at scale. Written from the terminal, not the whiteboard.
Rate limits, context compaction, session lifecycle management, and the hidden costs that only show up after thousands of real production runs. The operational reality of running Claude Code as infrastructure.
April 2026
12 min read
Why the hardest problems in production AI have nothing to do with model quality. The control plane became the product. Infrastructure, orchestration, and the unglamorous work that keeps autonomous systems running.
March 2026
15 min read
Multi-model routing, fallback chains, composite memory scoring, and what actually breaks when you push AI coding tools past their demo limits. From 5,252 production jobs across four model families.
March 2026
18 min read
The folder structure, routing logic, and real prompt patterns behind a production orchestration system. SQLite job queue, 28 skills, four model families, and the CLAUDE.md that ties it together.
March 2026
10 min read
A real Claude Code system running 500+ workers, multi-model review pipelines, LanceDB memory, and cron schedules. Built for daily production use, not a demo.
March 2026
10 min read
The infrastructure, job scheduling, and operational patterns behind running hundreds of concurrent AI workers in production without losing your mind.
April 2026
14 min read
A side-by-side comparison from daily use across all three tools. Performance, cost, privacy, and which one wins for different workflows.
April 2026
12 min read
Go deeper
IEEE-format studies on orchestration architecture, the autonomy tax, and memory systems. Same production data, more formal treatment.
See the infrastructure
Orchestration engine, worker pool, memory layers, tool integrations, and deployment topology.