What is agent ops?

Agent ops is the operational layer for running AI agents — spend control, consistency enforcement, visibility into what agents actually did, and the shared context they work from. Agents differ from plain LLM calls in kind, not degree: they run loops, hold growing context, use tools, and act with autonomy — so they need operations the way fleets need fleet management.

The cost dimension alone forces the discipline. An agent loop's cumulative input grows with the square of its length — a 20-step loop bills ~10× the tokens of its steps summed naively — and heavy agentic developers commonly run $400–1,500/month each. One ungoverned workflow can outspend a whole team.

For coding agents specifically, agent ops covers four questions:

• Spend — what does each repo/team/workflow cost, and why? (token ops)
• Consistency — does agent output follow the team's conventions, or drift per developer?
• Visibility — what did the agents actually do — which decisions, which files, what failed?
• Context — do all agents work from the same understanding of the codebase? (memory ops)

Agent ops is the gap between "every developer has an agent" and "the org can run agents safely at scale" — agents shipped as individual tools, and teams inherited the operations problem. That operational layer is what unerr is — see how it works across Claude Code, Cursor and Copilot, and the cost mechanics it governs.