Products / Builds

One rambled voice note. Two working apps.

Standfirst: Fable 5 was the first model where the speed and quality changed what’s worth building at all. Exhibit A: a taxi app for my kids that charges in chores.

Fable 5 was the first model that made me realise the ceiling had moved. Not “better benchmark scores” moved. Different-category-of-possible moved.

I didn’t give it a polished prompt. I rambled for about fifteen minutes into Wispr Flow, describing what I wanted. From that single messy description it did the planning first — worked out the core features, suggested surge pricing and the review system, wrote the copy, SEO-optimised the pages, designed the logos, and tested the flow before handing anything over.

It effectively ran my entire Hermes workflow — researcher, planner, builder, designer, copywriter, QA — in one pass, without step-by-step direction. That was the bit that blew my mind.

App one: Family Taxi

I spend a ridiculous amount of time as my kids’ Uber driver. School runs, dance, football, friends’ houses. Behind the wheel I kept circling the same thought: they have no real sense of what things cost. Not just money — time and effort.

So I built FamilyTaxi.uk.

The kids request a ride exactly like they would on Uber: pickup, destination, time. I set the fare — but it’s a chore, not money. Empty the dishwasher. Walk the dog. Mow the lawn. They can haggle by walking part of the way. Chore not done? The taxi doesn’t show up.

It has proper surge pricing — rain, early mornings, late nights and Sunday roast hours all push the chore cost up. Ratings run both ways: they rate the ride, I rate the chore, and everything goes on the permanent record. There’s a leaderboard.

It sounds like a joke. It works. They think ahead. They plan their week. They understand a ride has a real cost. And because it’s an app with ratings, reminders and a leaderboard, it’s a system — not another thing Dad nags about.

App two: RecipeMaster

Different frustration: I wanted a proper family recipe system — store recipes, plan meals, generate shopping lists — without paying £35 a month for tools that still felt half-finished.

So I built RecipeMaster. It pulls recipes from anywhere — Instagram, TikTok, websites, or typed in — and the AI works out the macros. Weekly planning knows exactly who’s eating each night, and keeps every meal within each person’s calorie target while respecting likes, dislikes and allergies. Diners score each meal afterwards. It knows what’s in the cupboards, so shopping lists only contain what’s needed. Invite-only, subscription-free, no bloat.

What actually changed

Twelve months ago, either app would have been a multi-week project I’d probably have abandoned. Now the marginal cost of a working household app is a voice note and a few evenings of review. Two things follow:

1. The backlog of “not worth it” ideas needs re-scoring. Things that never justified the effort suddenly do.
2. Judgement is the remaining bottleneck. When building is nearly free, knowing what to build — and what to leave unbuilt — is the entire job. Conveniently, that’s the product owner’s job description.

Next in the series: Why I make AI models argue before I spend a penny →