What AI can actually do for a small business right now

There's a lot of noise about AI right now, and most of it is either “this changes everything” or “this is all hype.” Neither is useful when you're running a business and trying to decide whether it's worth your time. So here's the grounded version, from people who use these tools every day to run an actual company.

The honest summary: AI is genuinely good at a narrow set of things right now, and genuinely bad at another set. The whole game is knowing which is which, and building your use around the line between them.

What it's genuinely good at today

The pattern to notice: AI is strong at the work around the work — the assembling, drafting, and sorting that surrounds a decision, so a human can get to the decision faster.

• First drafts. A rough version of an email, a service-page description, a set of FAQ answers, a job post. The blank page is the expensive part; getting a decent draft to react to and fix is where AI genuinely saves hours.
• Sorting and summarizing.Turning a pile of customer messages into categories, pulling the key points out of a long document, drafting a reply you then check. Tedious, repetitive reading — that's its home turf.
• Answering repeat questions. The same five questions your customers ask — hours, pricing ranges, service area, what to expect. A well-set-up assistant can handle the first pass on those, on your website, at 11pm.
• Getting unstuck.“Give me ten ways to phrase this,” “what am I forgetting on this checklist,” “explain this quote from my accountant in plain English.” A fast, patient thinking partner.

What it's genuinely bad at — and why

The limits aren't bugs that'll be patched next month. They come from what these tools are.

• Being right when it matters.AI can be confidently, fluently wrong. It'll invent a fact, a price, a policy, and state it in the same calm voice it uses for true things. That's fine when you're going to check the draft anyway. It's dangerous when the output goes straight to a customer.
• Being accountable.When it makes a promise to your customer, you own that promise. There's no “the computer said so.” That's exactly why the important stuff needs a human before it goes out the door.
• Judgment that needs context it doesn't have.It doesn't know your regulars, your margins, the customer who's one bad day from leaving a one-star review. It knows what's on the page in front of it. The wisdom is still yours.

The rule that keeps it useful

Here's how we run it ourselves, and what we'd tell any owner: let AI assemble the work, keep a human on every decision that reaches a customer. AI does the drafting and sorting; a person reviews, decides, and sends. Nothing that touches a customer goes out without a human looking at it first.

Where to actually start

Don't try to “adopt AI.” Pick one task that eats your time and doesn't touch a customer directly, and try it there for two weeks. Drafting your social posts. Summarizing the week's inquiries. Writing first-pass replies you then edit. Something low-stakes where a wrong answer costs you nothing but a redo.

If it saves you real time, expand carefully. If it doesn't, you've lost two weeks of light experimenting instead of a big software bill and a workflow you resent. That's the honest way in — small, checkable, and always with you making the calls that count.