AI agents for small business: what works and what is still hype

Every week a client asks whether they should be "doing something with AI". The honest answer is that some things work extremely well right now and some are still demos that fall apart on contact with real customers.

What works today

Answering questions from your own documents. Point a model at your manuals, contracts and past tickets and it will answer accurately. This is the single most reliable business use of AI right now, and it saves real hours.

Classifying and routing. Reading incoming messages and deciding what they are about, who should handle them and how urgent they are. Boring, invisible, and it works almost perfectly.

Drafting from structure. Turning a set of facts into a first draft — proposals, reports, summaries. A human still reviews, but from a draft instead of a blank page.

What is still hype

Fully autonomous customer service. It handles the easy 60% well and then confidently invents something wrong on the case that matters. Use it as a first filter with a clear route to a human, not as a replacement.

Agents that "run your business". The demos are spectacular. In production they need so much supervision that the supervision costs more than the work.

AI that decides. Anything where a wrong answer has real consequences — pricing, contracts, compliance — still needs a person accountable at the end.

The businesses getting value from AI are not the ones doing the most ambitious things. They are the ones who picked one boring, repetitive process and did it properly.

Where to start

Pick the question your team answers most often. Build something that answers it from your own documents. Measure the hours saved for a month. That single project will teach you more about where AI fits in your business than any amount of reading.