Blog
Writing on reliability, AI and production.
Essays and engineering notes from the team building Dalton: how production fails, how AI changes it, and what we learn running agents in real systems.
What we learn building Dalton.
From Signals to Causes: What AI Investigation Needs
Signals tell you that something moved, but rarely why. Try to find the cause of one incident from its signals alone, then take three questions into your next vendor call.
EngineeringWe Benchmarked Jev Against Our Own Classifiers. Every Router Stalled at 70%.
We set out to replace five LLM classifiers with a model built for the job. It was 15x faster and up to 44x cheaper - and no more accurate. Tracing all 507 decisions showed why, and the answer was sixteen text files none of us had read.
EngineeringEight Rules for Benchmarking a Model You're About to Trust
One request, ten runs, two defensible answers eleven seconds apart. Every rule below is here because skipping it would have let us publish something flattering and wrong - including the one that nearly made us reject a model that worked.
ManifestoThe Double Exposure: When Your AI Agents Run on AI-Generated Code
In March 2026, Amazon lost 6.3 million orders in a single afternoon. The cause wasn't an AI agent. It wasn't AI-generated code either. It was both - running on top of each other. This is the failure surface nobody's measuring yet.
InnovationMythos: What Anthropic's Own Reliability Team Says - and What It Means for Production AI
Anthropic's model card includes a candid reliability assessment of Claude Mythos. We break down what their own engineers found - where it's a step change, where it still falls short, and what this means for teams running AI in production.
EngineeringThe Three Meanings of "Nothing" in AI Agent Tool-Calling
Mid-investigation, a tool call comes back empty. One agent retries endlessly. Another reports a failure. A third keeps searching across every data source it can find. Same empty response, three completely different problems.
IndustryWhat is AI SRE? The History, the Market, and What Comes Next
AI SRE did not appear out of nowhere. Here's where it came from, how the market is forming, and why the next phase will be about prevention, not prettier incident response.
InnovationAI Is Writing 46% of Your Code. Nobody's Watching What It Breaks.
We use AI coding tools every day. They're the future. But 46% AI-generated code, 98% more PRs, and 30% more change failures is what that future actually looks like - and nobody's talking about it honestly.
AIYour AI Strategy Is a Data Strategy Wearing a Costume
Most companies think they have an AI problem. They don't. They have a data problem - messy pipelines, no context, garbage semantics - and they're throwing models at it hoping nobody notices.
ManifestoPostmortem: Reactive Operations
We automated detection twenty years ago (to some extent...) - Investigation is still manual. Here's why that's a problem.