Building effective agents
Anthropic Engineering
Why it matters: A durable architecture reference for deciding when workflows are enough and when agent autonomy is actually warranted.
Read source ↗Documenting the operational journey from AI learner to systems architect — through public research, production builds, and leverage-focused engineering.
Operational AI infrastructure being designed, shipped, and rewritten in public. Status reflects the latest published system note.
Full systems index →Curriculum graph, capability ledger, spaced practice loop.
Hybrid retrieval with reranking. Embedding-store migration tracked as an active infrastructure change.
Trace-backed regression detection across all shipped agents.
Notebook → outline → draft → review → publish, traced end-to-end.
From learner to operator to architect — a deliberate progression model for technical depth.
Designing systems, content, and careers that compound without proportional input.
Mapping technical work to economic leverage and durable, defensible outcomes.
The capability ladder for production AI: evals, reliability, observability, ownership.
A second-brain operating system for converting reading and building into compounding artifacts.
Architecture iterations, post-mortems, and shipped changes — each entry annotated with the lesson it earned.
All logs →Migrated lesson dependencies from a flat list to a directed graph. Surfaced 23 cycles. Notes on resolution heuristics.
Switched from naive cosine over a single index to hybrid BM25 + vector with reranking. Recall up 28%, latency flat.
Schema drift in upstream traces silently dropped 11% of eval runs for two weeks. Detection plan + guardrails added.
Pair-programmed with Claude on a 4k-line module split. Documented prompts, failure modes, and human-review pattern.
A deliberately small reading layer: external work selected for the questions it changes, not for the volume of the feed.
Anthropic Engineering
Why it matters: A durable architecture reference for deciding when workflows are enough and when agent autonomy is actually warranted.
Read source ↗OpenAI
Why it matters: Useful framing for turning fuzzy quality expectations into explicit, measurable evaluation loops.
Read source ↗Anthropic Research
Why it matters: Connects agent autonomy to human control, transparency, security, and operational boundaries.
Read source ↗