After 100 meetups as London DevOps, we continue in our refreshed identity with London DevOps AI meetup #002 at Civo Tech Junction. We will have two talks on getting real value and real safety from AI coding agents. Rob Coward will argue that agents should loop against a human-written specification rather than choose the approach themselves. Ace Vinayak will then share how his team built fail-closed, deterministic guardrails around every agent tool call. And there will be drinks and pizza as usual too. **Schedule** 6:00pm - Arrival 6:45pm - Introductions 7:00pm - The Talks **From Do-While to Loop-Until: Spec-Driven Delivery with AI Agents (Rob Coward)** Coding agents are brilliant at writing code and terrible at delivering features. The usual loop: hand the agent a ticket, let it choose the approach, review the PR at the end; leaves you reverse-engineering a finished feature, judging "good enough" without knowing what was rejected or why. The agent took the interesting decisions and left you the chore. Invert it. You make the engineering trade-offs and write the spec; the agent loops until the code matches it. Review moves to the spec, where a rejection costs minutes instead of hours, and where the \*how\* stays human. Rob Coward is a Senior DevOps Consultant. **The Agent that Dropped the Table: Fail-Closed Guardrails for AI Coding Agents (Ace Vinayak)** The DevOps story: coding agents now run shell commands, edit CI config, push to git and call cloud APIs, usually with an engineer's own credentials. We tried the usual controls (sandboxes, command allowlists, hooks, a second LLM reviewing the first) and each broke in its own way. So we built a gate in front of every tool call: it breaks each action into facts, and decides it against written policy, deterministically, failing closed when it isn't sure. I'll cover what we got wrong, the bypasses we found in our own gate, how allow / ask / block tiers keep developers moving instead of clicking "approve" all day, and tre

Description supplied by the event organiser.