Field notes from production.
Engineering posts on the things we hit and the patterns we now reach for. No predictions. No trend pieces.
- // engineering · 2026 q1LangGraphMulti-AgentArchitecture
2026-02-04 · 11 min
Four LangGraph topology patterns that survive production.
We've shipped 14 LangGraph topologies in production. Four patterns showed up everywhere; one only worked in dev. The field-tested set.
read → - // open source · 2025 q4LangOpsOpen SourceBenchmarks
2025-12-09 · 7 min
Introducing guardrail-eval-bench: a deterministic eval set for prompt-injection detectors.
Most prompt-injection benchmarks are LLM-judged — the bug they claim to detect. We're releasing a deterministic eval set with 1,800 hand-labeled cases.
read → - // case notes · 2025 q3Case NotesInsuranceGuardrails
2025-10-22 · 14 min
What we learned migrating a $2.4B claims-intake queue to an agentic pipeline.
Insurance regulators don't accept 'the model said so' as a control. How we structured a multi-agent FNOL pipeline an audit team could read.
read → - // engineering · 2026 q2LangGraphReliabilityLong-Horizon
2026-04-18 · 12 min
What stays stable past hour eight: long-horizon agent runs in production.
Agents that run for hours fail differently than agents that run for seconds. Five patterns for surviving token flushes, outages, and model upgrades.
read →
// rss feed planned · subscribe via the intake agent if you want a note when a new post lands.