AI tools that are actually useful vs. the ones that are just hype

Every app now has AI in it the way every snack once had "natural" on the label — as an incantation, not an ingredient. This desk's AI coverage has one house rule: no theatre. So here's the honest sort, offered as opinion and analysis, of which AI tools change what you can do and which just changed their own pricing page.

The five that earn their keep

1. Drafting partner. The assistant that turns your bullet points into a first draft — the single highest-value everyday use, because the draft exists to be argued with, and arguing is what the models do well (any of the big three does this; the differences are editing time). 2. Code help, honestly used. Explaining an unfamiliar error, writing the boilerplate you've written forty times, reviewing a diff. The failure mode — trusting generated code you can't read — is a you-problem the tool can't fix (the error-message skill still comes first). 3. The summariser with a verification habit. Long documents to a working summary is a genuine superpower — with the rule that anything you'll act on or repeat gets checked at its source, because summarisers are confident about details. 4. The learning partner. "Explain this to me three ways, then quiz me" is the best free tutor ever shipped — and the one use where the mistakes don't matter, because you're checking against the course material anyway. 5. Data janitor. Reformatting, extracting, converting — the tedious middle of office life, done in seconds with a paste and one instruction.

The four that are mostly label

The "AI-powered" note-taker that summarises your meetings: the summary is real, the promise that you'll read it isn't — it's a transcript with extra steps unless it changes what you do after. AI photo "enhancers" that hallucinate detail into your images: impressive demos, uncanny outputs, unusable for anything you care about. Chatbots bolted onto support pages to deflect tickets rather than answer questions — you've met them; you've hated them. And the AI-feature upcharge: the app you pay for that now wants a further fee for the same product with a chat box — the free tiers usually cover the same need, which tells you what the feature actually costs.

The one-minute test

When a new AI tool crosses your desk, ask three things. One: does it give me a draft I can check, or an answer I have to trust? (Drafts are useful; trust-me-answers are how errors get published.) Two: what does it cost when it's wrong? (Learning partner: nothing. Legal brief: everything.) Three: is the AI doing the task or decorating it? Strip the label and ask what remains — if the answer is "a slower app with a chat box", you've met the hype category. The tools that pass all three are already on this desk's AI shelf, re-opened on every republish — because this is the one category where even honest reviews have an expiry date.

Opinion and analysis, offered as such · reviewed September 2026.