Agents in ProductionAgents in Production

Through 2023 and 2024, the dominant AI product was a chatbot: you asked, it answered, and a human did everything else. By 2025 the frame had shifted to agents, systems that take a goal, plan a sequence of steps, call tools, react to results, and keep going until the task is done or something stops them. The industry widely declared it the year of the agent.

The ingredients had matured together. Models trained to reason before acting made multi-step plans less brittle, standards like the Model Context Protocol gave agents a uniform way to reach tools and data, and benchmarks like SWE-bench and GAIA measured whether systems could resolve real GitHub issues or complete realistic assistant tasks, not just answer trivia.

The demos were intoxicating: an agent takes a bug report, reproduces the issue, writes a fix, runs the tests, and opens a pull request. But teams that moved from demo to deployment learned the same lesson everywhere. A step that succeeds 95 percent of the time fails more than one run in three when chained twenty steps deep, and errors compound silently unless something catches them.

So production agent engineering became, above all, failure engineering. Real systems earned their keep through checkpoints and retries, permission boundaries on what an agent may touch, human approval gates for consequential actions, evaluation suites run like regression tests, and traces that let engineers reconstruct why an agent did what it did.

The successes were narrower than the hype but real: coding agents resolving routine issues, support agents executing refunds and account changes end to end, research agents assembling cited briefs. The pattern favored constrained domains with verifiable outcomes over open-ended autonomy.

The moment marked a quiet redefinition of what AI deployment means. The question stopped being how smart the model was and became how well the system around it handled the model being wrong, and that skill, unglamorous and operational, separated the teams whose agents shipped from those whose agents stayed demos.