What they built

Concord — an MCP that turns one command into a cross-department check

Most "AI DevOps" demos are chatbots wearing a hard hat. Concord actually runs the pipeline — commit, build, cost-check, canary deploy, monitor, loop — with nobody touching it. It watches five clusters across three regions, live security threats, and real department budgets at the same time, and it doesn't just react, it does the math first: what does halting this cluster actually cost, whose contract does it break, whose budget goes negative. If the answer crosses a hard line, it refuses to act — even fully autonomous, it won't drain a team's budget below its safety floor. Everything else, it just handles. When a human does need to step in, it hands over a real briefing — financial exposure, conflict type, a live AI-generated root-cause chain — with an SLA clock ticking, not a wall of logs to decode. Every decision, human or machine, gets hash-chained into a tamper-evident audit trail. Built on Nitrostack/MCP, the same brain runs as a live dashboard and as an MCP server any AI client can call — same rules, same receipts, either way.