Firstly to clarify what I mean with AI agent: A script which utilizes an LLM is not an agent. It's automation, alternatively a "workflow" (according to Anthropic). An AI agent is a system which is handling complex semantic input by gathering information about the input via tooling + memory and then lastly process it into some "judgement based" response.

The shell script which writes a poem about the current weather in your city -> automation.

A chatbot for customer support -> agent. Codex/Claude Code -> agent.

Primary difference being that an agent build its own context depending on the task that it needs to solve.

I've heard a lot of buzz around AI agents, but I've yet to see that many success stories where they fit nicely and provide more value than what they cost to develop and maintain (exception being coding agents). So, please share your AI agent success stories!