Topic
- Screening candidates with an LLM without building a discrimination engineHiring is the use case where an unexamined model does the most damage, and where the damage is hardest to see from the inside. The engineering controls matter more here than anywhere else.2 min read· HR· Regulated· AI in production
- Voice agents: the latency budget is the designIn text, a two-second pause is imperceptible. In conversation it is the difference between a system that feels present and one people talk over. Almost every voice architecture decision follows from that.2 min read· Voice· Agent architecture· AI in production
- Model upgrades are breaking changesA new model version is a dependency upgrade to the least deterministic component in the system, and it usually arrives without a changelog that describes what will change about your behaviour.2 min read· LLM workflows· Delivery· AI in production
- Observability for LLM systems: what to log when the output is textConventional monitoring tells you the request succeeded. For a model-backed system that is nearly useless — the interesting failure returns 200 and says something wrong.2 min read· Observability· Platform· AI in production
- PaperEvaluating agents in productionA working method for deciding whether a change made an agent better or worse — how to build the set, what to measure, how to gate a release, and how to keep it honest as production moves.4 min read· Evaluation· AI in production
- AI in regulated workflows: what changes when a human must sign offA required human approval is often treated as a compliance formality bolted on at the end. Designed properly it is the most useful architectural constraint you will get.2 min read· Regulated· Agent architecture· AI in production
- What “production” means for an agentEleven questions that separate a system from a demo. None of them are about model quality, and most projects that stall have not answered more than three.3 min read· AI in production· Platform
- Tool permissions are an authorisation problem, not a prompt problem"You must never issue a refund over £500" in a system prompt is not a control. It is a request. The boundary belongs in the tool layer, where it can be enforced.3 min read· Agent architecture· Security· AI in production
- Why your evaluation suite passes and production still breaksAn eval set is a sample from a distribution you chose. Production is a sample from one you did not. Almost every surprising failure lives in that gap.3 min read· Evaluation· AI in production
