Articles, one at a time.
Every piece cites primary sources and adds something you can act on: a worked example, a tested command, a documented limitation. Drafts that only summarize what already exists stay unpublished.
Agent Reliability Engineering: Retries, Loop Detection, and Timeout Budgets for Production LLM Agents
Hands-on testing of retry with exponential backoff, loop detection via step-count limits, timeout budgets, and human escalation triggers for production LLM agents, with a ReliabilityBench stress-test harness.
Read →
LangGraph in Production: Checkpointing, Postgres Persistence, and the Silent Re-Execution Trap We Hit Firsthand
We deployed LangGraph with a Postgres checkpointer in our lab, reproduced the silent re-execution trap on long tool calls, and documented the exact gotchas, workarounds, and cost math for production agent deployments.
Read →
Qdrant Hybrid Search in Production: Our Dense + Sparse + RRF Pipeline, Tested, Broken, and Fixed
We ran Qdrant's dense+sparse hybrid search with FastEmbed and RRF fusion end-to-end on a single Docker container, hit every documented failure mode, and turned the production checklist into fixes we actually applied.
Read →
Google Antigravity 2.0: Can Its Browser-Driving Agents Survive Real Repo Work?
We installed Google Antigravity 2.0 in our lab and put its Agent Manager and browser-driving agents through a scripted battery of real repo tasks, including a browser-driven E2E test. Here is what worked, what broke, and whether it justifies a seat.
Read →
From Notebook to Production SLA: Running vLLM on Kubernetes with the Production Stack
A practitioner's guide to moving vLLM from a single-GPU notebook demo to a Kubernetes deployment with measured p50/p95 latency under concurrency, grounded in the vLLM production-stack release and the 2026 GLM-5.2 production SLA architecture.
Read →
Your Agent Is Not a User: Giving AI Agents Their Own OAuth Identity with Scoped, Revocable Credentials
Why sharing your users' OAuth tokens with AI agents breaks audit, least privilege, and enterprise deals — and how to run agent identity with OBO exchange on a local stack.
Read →
Sonnet 5 Pricing: Why a Cheaper Token Can Cost You More
Sonnet 5 costs a third less per token and counts your text differently. The break-even math, plus the caching and routing architecture that decides it.
Read →
Your LLM Bill Is an Architecture Problem, Not a Prompt Problem
Provider invoices bill credentials, not customers. Three LLM cost attribution patterns, a list-price model, and which control to build first.
Read →
Semantic Caching vs. Prompt Caching: Where Break-Even Sits
Two caching layers on one workload, priced from Anthropic's official cache rates. Where each pays, and the measurement that tells you which fits.
Read →
Supabase Removes logs.all on 2026-09-23: Port Your Queries First
Supabase kills the logs.all endpoint on 2026-09-23 and the replacement speaks ClickHouse SQL only. The inventory scan and query rewrites to ship first.
Read →
OpenAI Kills the Videos API With No Successor: A 4-Week Exit Plan
The Videos API and every sora-2 model shut down 2026-09-24 with an empty replacement column. The inventory scan and exit matrix to decide before the date.
Read →
Your Coding Agent's Allowlist Is Not a Sandbox
An allowlist approves names, not behaviors. Why environment poisoning turns a trusted command into code execution, and a config auditor that finds it.
Read →