IDEAS, DECISIONS & LESSONS
Notes on AI, architecture and building useful products.
155 articles
30 Sep 2026 · 5 min read
architecture
Worker Previews spin up an isolated Durable Object namespace and Container per branch with one wrangler command. The service-binding gap is the part worth reading before you wire this into CI.
Read more30 Sep 2026 · 5 min read
architecture
Embed 5 Pro and Fast share one vector space: index once with Pro, query with either model, lose 1.6-2.7% retrieval quality and gain 2.4x query throughput. Here's the actual math for when that trade is worth making.
Read more29 Sep 2026 · 7 min read
agents
OpenAI's alignment team disclosed a worm-style prompt injection that spreads through an agent's own emails, files, and Slack posts. Most teams only validate input. That's not enough anymore.
Read more29 Sep 2026 · 5 min read
architecture
Uber's error-ownership pattern fixes a bug hiding in almost every service mesh: retries fire at every hop, even when the hop didn't cause the failure. The retry-amplification math explains why, and it's a five-line middleware fix.
Read more29 Sep 2026 · 5 min read
agents
Sonnet 5.5 ships at the same $/token price as Sonnet 5 but does the same work in fewer steps. If your cost dashboard only tracks tokens, you'll never see the win — or the next regression.
Read more29 Sep 2026 · 4 min read
ai
Public Preview lets you tag an existing OpenAPI spec and expose it to AI agents at /mcp, reusing your current auth and quotas. Here's the real config, the real limits, and why the tool description field is now the most important line in your spec.
Read more27 Sep 2026 · 4 min read
agents
Anthropic just signed a 7-year, up-to-$20.6B deal with Akamai, with a real equity warrant attached. Reading the deal alongside Akamai's own latency argument, and testing whether edge compute actually fixes agentic tool-call latency.
Read more24 Sep 2026 · 4 min read
agents
Strands Harness ships with hard numbers on context pruning: truncate tool results over 1,500 tokens, summarize past 85% context usage. 28% lower token cost, 77% cheaper than Claude Code on Terminal-Bench 2.1. What that means if you already hand-rolled this yourself.
Read more24 Sep 2026 · 4 min read
ai
Cloudflare's Python Workers went GA with real TCP sockets — not a REST-only sandbox. aiomysql, asyncpg, FastAPI, and yes, openai and langchain now run natively at the edge. A look at what actually changed under the hood, and where I'd still say no.
Read more
