Paul TinkerProduct designer · Design systems Back to portfolio

AI-native products & design systems

A working log of designing and building digital products with AI

An idea I'd had for years, taken all the way to Google Play: app, site, brand and backend. I designed it and directed the build, and the first time the whole product was mine.

Edgey home screen: the logo above a four-tile puzzle that solves itself, with the three rules.
HomeThree rules and a small demo board. One tile keeps turning out of place and back, so you watch its edges break and snap shut again. That shows the whole game without a tutorial.
A board mid-solve, with matched seams capped in colour.
A board, mid-solveDrag a tile onto another to swap; tap to turn it a quarter. Matched seams cap in colour, so progress shows without a score.
The levels screen, showing the shapes of locked packs.
Every level, visibleLocked packs still show their real shapes. Boards growing from 3×3 to 6×6 are the argument for paying; before this they were bought unseen.
The leaderboard, ranked on fewest moves, with generated names and animal avatars.
Ranked on fewest movesNo account, no sign-up, no email. A name from a word list and an animal, so there's no free-text field anywhere and nothing to moderate.
Trading coins in to drop one tile into place.
CoinsEarned by solving, spent to drop one tile into place: ten coins for one tile, and your score still counts.
The player screen: choose one of fifty-six animals and reroll your name.
Who you areFifty-six animals, and a name you can reroll. The avatar travels to the server as a number, never as text, which keeps the privacy policy true.
The Edgey app icon, a lightning bolt with four coloured pills, above the cream EDGEY wordmark on the game's teal background.
Logo and app iconThe EDGEY wordmark and the launcher icon that shipped.
1 / 7

The question this time: when building takes a day, where does the design work go? Edgey, a puzzle game I'd wanted to make for years, was the test.

Built entirely by prompting. I didn't write a line of code or open an editor. Engine, backend, build tooling and site were all made in Claude Code by describing what I wanted and reviewing what came back.

Knowing which surfaces earn original work is most of what design judgement is when the whole build takes a day.

QA turned out to be the job. Anything I could write down as a rule got checked by machine, exhaustively: every board verified solvable, the check that stops fake scores reaching the leaderboard tested on every level rather than assumed to work. What was left needed someone to look at it: titles sitting 35px off-centre, store screenshots rendered at desktop width, a locked-level button that was dead on a slow connection. None of those came from reasoning about code.

The concept and the design were mine. An edge-matching puzzle I'd written down years ago and never built: every tile carries four marks, every touching pair must agree. Tile marks, palette, icon, store screenshots. Whether you can tell at a glance that two edges match comes down to how the marks are drawn. That's design work, not code. The avatars and game icons are off-the-shelf emoji, deliberately. Knowing which surfaces earn original work is most of what design judgement is when the whole build takes a day.

App launcher iterations

First icon: one tile with four emoji marks on teal.
01 · Where it startedThe game drawn honestly: one tile, four marks. At 48px it is four smudges inside two frames.
Second icon: a single yellow bolt on a dark tile.
02 · One markLegible at any size, and it says nothing. A yellow bolt could be any app in the drawer.
Third icon: two tiles meeting at a pink seam.
03 · The seam aloneThe mechanic as pure shape. Striking large, and at 48px it tells a new player nothing at all.
Fourth icon: the bolt with coloured pills on all four edges.
04 · Pills on the edgesFour edges, four marks: back, but as colour rather than detail. Still two frames around a small picture.
The shipped icon: the bolt and four coloured pills, full bleed on near-black.
05 · ShippedFull bleed on near-black. The tile and its teal surround gone, the mark given the whole canvas, each pill in a pool of its own colour.
1 / 5

Also

  • Quick user testing: a few people played it, suggested a mechanic that made it into the game, broke a rule I hadn't thought through, and found bugs I'd have shipped. The fix for the broken rule was a better rule, not an exception
  • Signing keys, and the setup inside my own Google Play and payments accounts, I handled myself, so nothing sensitive went through the model
  • A web view inside a native shell, the same approach as a great many shipped apps. Worth saying rather than leaving someone to find out

Not yet. Live on Google Play, but the iPhone version is still in review with Apple.

Still on the list. Versioning and visual regression, from 21 September, are both untouched. They're the two that decide whether any of this holds up with other people rather than alone.

In one week: two design systems taken into code and Storybook, and a game submitted to the Play Store. The bottleneck has moved. It isn't building any more, it's knowing what's worth building and noticing when what came back is subtly not right. At £2.49 a copy, if it makes any money it'll pay for the coffee that built it.

Get it on Google Play