What should you actually turn into an AI agent?

Founders are building agents too early. The hard question is no longer whether AI can automate a task. In many cases, it can. The harder question is whether that task is worth automating at all — and whether the right answer is a prompt, a workflow, a skill, or a true autonomous agent.

The agent boom has created a strange new form of startup waste: automating work because it looks automatable. A founder sees a repetitive task, opens an agent builder, wires together five tools, and feels productive. A month later the agent saves 25 minutes a week, breaks whenever the workflow changes, and still needs someone to watch it.

That is not leverage. It is a small software product you now have to maintain.

The useful question is economic: if this task disappeared, became faster, or became partially autonomous, would the startup materially improve?

The goal is not to build more agents. The goal is to buy back the most valuable human time at a sensible cost and risk.