What 'AI agents' actually mean for a business (and what they don't)

“AI agent” has become the kind of phrase that means everything and therefore nothing. Before you spend money on one, it's worth being precise about what the word should mean — and what it shouldn't.

A useful definition

An agent is a system that can take a goal, break it into steps, use tools to act on each step, observe the result, and adjust. The important part isn't the language model — it's the loop: reason, act, check, repeat. A chatbot answers a question. An agent does a job.

Where agents genuinely earn their keep

• •Reading messy documents and extracting structured data reliably.
• •Multi-step tasks that need judgment at each hop — triage, routing, drafting.
• •Pulling together information that lives across several systems to answer one question.

Where they're the wrong tool

If a task is deterministic — same input, same correct output, every time — you don't want an agent, you want code. Agents add value precisely where judgment and ambiguity live. Pointing one at a problem a simple script solves is slower, costlier, and less predictable. Good engineering is knowing which is which.

The honest version of the pitch

Agents are real capability, not a magic wand. Built well — on a modern stack, with the right guardrails and a human in the loop where it matters — they take real work off real people. Built badly, they're an expensive way to be wrong confidently. The difference is engineering, which is the whole reason to work with engineers.