Durable Execution for AI Agents: Step Limits, Replay Traps

Durable execution for AI agents explained: why tool calls still run twice, step and history caps in Temporal, Inngest and Step Functions, and the fix.

Durable execution for AI agents explained: why tool calls still run twice, step and history caps in Temporal, Inngest and Step Functions, and the fix.

Decision model vs LLM: what Amazon's Strands Decider swap really saved, the break-even volume for a GPU endpoint, and a safe Node.js shadow-mode rollout.

AlphaProof Code used AI agents to prove Signal's compiled code secure in Lean. See the numbers, the limits, and a gatekeeper pattern for Node.js agents.

Cancel LLM stream on client disconnect in Node.js: why req.on('close') fails, the res.writableFinished fix, plus OpenAI, Claude and Vercel code samples.

pgvector vs MongoDB Atlas Vector Search for Node.js: filtering, consistency, dimension limits and hybrid search compared from the official docs, with code.

Agent memory architecture explained: index limits, one page per entity and path checks, with a tested Python store and Anthropic's Nov 2026 findings.

LLM structured outputs stop AI APIs from returning broken JSON. Learn the Node.js + Zod fix for OpenAI and Claude, plus refusal and truncation gotchas.

GPTBot trains OpenAI's models; OAI-SearchBot powers ChatGPT search. Learn which AI crawlers to block, with a tested robots.txt template for 2026.

AI agent sandbox lessons from OpenAI's 18,000-edit wiki breakout: why GET-only egress fails, how /etc/hosts beat NO_PROXY, and the fixes to ship now.