Ideabro Machine Pitches Ten AI Startup Ideas, and a Human Grades Every One

The brief

We gave Ideabro Machine, our synthetic correspondent, a one-line brief: "Pitch AI startup ideas that solve boring, real problems for ordinary people, not for other startups." It produced twelve. We cut two that closely resembled products already on the market and kept ten.

Then Theo Brooks, who covers startups for us, graded each one. The pitches below are the machine's, lightly trimmed. The verdicts, in italics, are Theo's.

The ten pitches

1. Warranty Butler. An agent that reads your email receipts, tracks every warranty and return window, and files claims when something breaks. Photograph the broken kettle and it does the paperwork.

Verdict: Survives. It's boring and valuable, and it gets more useful the longer you use it. The hard part is getting into inboxes safely, so earn trust slowly.

2. FridgeCount. Snap a photo of your fridge each morning and get meal plans that use whatever is about to go off.

Verdict: Dead. Every version of this dies of daily-photo fatigue. Nobody photographs their fridge for more than a week.

3. QuoteCheck. Homeowners upload quotes from builders or plumbers and get a plain-language breakdown, line-by-line comparisons and questions to ask before signing.

Verdict: Survives, with a caveat. People are desperate for this, but prices vary wildly by region, and without local pricing data it's a confident guess. Start in one city.

4. Afterwards. A careful assistant for the paperwork after a death: closing accounts, notifying organisations, and tracking what's been done and what's left.

Verdict: Survives, if it's handled with extraordinary care. This is real pain at the worst moment of people's lives. A human must stay in the loop and the tone has to be perfect. It won't be a growth-hack business, but it could be a beautiful one.

5. Tuck. Personalised bedtime stories starring your child, generated nightly.

Verdict: Dead. It's crowded and cheap to copy, and many parents would rather read the same three books forever. Kids like repetition more than novelty.

6. Slangbuddy. An AI pen pal for language learners that teaches local slang and texting habits, not textbook phrases.

Verdict: A feature, not a company. It's a lovely idea that the big language-learning apps could add in a sprint.

7. ClearMonday. An agent that negotiates your calendar to keep Mondays meeting-free.

Verdict: Dead. Calendars aren't a scheduling problem; they're a politics problem. No agent wins an argument with your boss's assistant.

8. Reverse Recruit. A job board where AI agents apply to humans. You describe your skills, and companies' agents pitch you roles.

Verdict: Fun headline, no business. Recruiters already flood inboxes, and this adds more noise with extra steps.

9. RotaKind. A scheduling agent for small volunteer groups like food banks, sports clubs and community gardens, which handles swaps, reminders and no-shows by text.

Verdict: Maybe. The pain is real and the users would be very loyal, but there's almost no budget. It could work as a cheap tool funded by a few larger charities, or as a side project that does a lot of good.

10. LeaseLight. Upload a rental agreement and get a plain-language summary with unusual clauses flagged.

Verdict: Maybe. Renters need it, and the liability is serious. It works best as a free tool backed by tenant organisations, with clear "this is not legal advice" boundaries and a route to a human adviser.

What survived

Three ideas got a clear yes (Warranty Butler, QuoteCheck and Afterwards), and they have something in common. Each one builds on paperwork people already have, solves a problem they feel often or intensely, and gets better as it builds up history. None of them is a toy.

Two got a maybe. Both serve people with real needs and little money, which is often where the most useful software never gets built.

Editor's note: what the machine gets wrong about startups

Theo Brooks writes: Reading all twelve pitches together was instructive. Ideabro Machine is very good at spotting needs and weaker at seeing the things that kill companies.

• It underestimates habit. FridgeCount and Tuck assume people will change daily routines for a slightly better outcome. Mostly they won't.
• It ignores distribution. Almost no pitch explained how the first hundred customers would find the product.
• It doesn't see the bundle. Slangbuddy would make a good feature for an app that already has millions of learners.
• It's kind. Not one pitch tackled a problem people would be embarrassed to admit to, and that's often where the money is.

That last point might be a strength. As a brainstorming partner the machine is generous, fast and tireless. As a judge, it needs a human with some scar tissue, which is why we'll keep this format.