Letting Go of Control: Building a Game with My Kids and AI

I’m a freelance developer who wanted to combine making games with spending time with my children. Naturally, I added AI to make things simpler. That went about as smoothly as you’d expect.

This talk follows the experiments, failed attempts and changes of direction behind Don’t Go To Bed, a game we built together for Android and iOS using Kotlin and Compose Multiplatform.

I share where AI helped, where it created more work, how playable prototypes changed our decisions, and what happened when I stopped treating my children’s ideas like requirements I could quietly rewrite.

For developers curious about AI-assisted development, side projects and getting useful feedback before building too much.

These slides accompany a live talk. Videos and animations appear as still images in this PDF.

Game and project links: https://jamescullimore.dev/dontgotobed/

Transcript

1. THE OLD DREAM Building a Game with My Kids and AI JAMES CULLIMORE 01
2. THE OLD DREAM It's just a story. Maybe it'll help you try yours. 02
3. THE OLD DREAM 10 “I want to make a game.” YEARS OLD 03
4. THE OLD DREAM Games were always there. PS2. My brother Tom. Ego intact. 2003 04
5. THE OLD DREAM Eventually, I built it. Elchem · Droidcon London, 2025 No engine. No budget. No clue. Watch last year’s talk jamescullimore.dev/elchem/ 05
6. THE OLD DREAM Goal achieved. What do we do now? 06
7. THE OLD DREAM Become brutally unhappy. 07
8. THE OLD DREAM Find another adventure. This time, together. 08
9. THE BOYS INVENT A GAME 02 The boys take over. Their imagination. Our next adventure. 09
10. THE BOYS INVENT A GAME They love this as much as I do. 10
11. THE BOYS INVENT A GAME Boys, would yous like to develop your own game? 11
12. THE BOYS INVENT A GAME YES. 12
13. THE BOYS INVENT A GAME Leo grabs the scrapbook. The first design document. NO TICKETS REQUIRED 13
14. THE BOYS INVENT A GAME Three halves of a game. ½ ½ ½ Minecraft Stunt Car Extreme Mario Kart Ambition: excellent. Maths: negotiable. 14
15. THE BOYS INVENT A GAME MULTIPLAYER They wanted to play together. I started thinking about servers. 15
16. THE BOYS INVENT A GAME I am one developer. Maybe start a little smaller. 16
17. THE BOYS INVENT A GAME Meet the development team. LEO IAN DADDY Ideas + drumsticks Chaos + rabbit Implementation + fists 17
18. THE BOYS INVENT A GAME Find Mammy. Before bed. Ian wanted Mammy. That became the story. THE ADVENTURE US FIND MAMMY 18
19. THE BOYS INVENT A GAME And then the ideas kept coming. Original scrapbook ideas 01 02 03 Drum-kit powers A giant cat Berserker Leo Ian Daddy Plus bosses. Biomes. Crafting. Treasure. Multiplayer. 19
20. THE ORIGINAL PLAN FAILS 03 Let's build it. The obvious plan lasted about ten minutes. 20
21. THE ORIGINAL PLAN FAILS Ten minutes later… 21
22. THE ORIGINAL PLAN FAILS They wanted to create. I started coding. I needed to give them something to do. 22
23. THE ORIGINAL PLAN FAILS Make something they can see. Now we all had something to look at. Try a look. Change a detail. An opinion becomes a visible change. 23
24. THE ORIGINAL PLAN FAILS A face. A pose. A walk. FACE TURN MOVE 24
25. THE ORIGINAL PLAN FAILS Now we were making something together. Decide. See it. Laugh. Change it. 25
26. THE ORIGINAL PLAN FAILS Somebody still has to write the code. 26
27. AI AND ANDROID 04 Let an agent try. An experiment in keeping the fun alive. 27
28. AI AND ANDROID I didn't trust it. Especially the complicated bits. 28
29. AI AND ANDROID One more problem. iPads. The devices they actually use. ? How do we get there? Android → iPad 29
30. AI AND ANDROID Three unknowns at once. The first KMP + Junie attempt did not go well. ? ? ? KMP / CMP The agent Me New platform work New way of building New things to learn 30
31. AI AND ANDROID Back to Android. Back on a platform I knew. 01 02 03 Android Junie A playable game Familiar platform New workflow Something to test 31
32. AI AND ANDROID Jump. Land. Play. One small mechanic was enough to change the mood. var v = 1980f val g = -3000f val frameMs = 16L val dt = frameMs / 1000f v += g * dt heightPx += v * dt fdbaae2 · MainActivity.kt · selected lines; comments and surrounding control flow omitted 32
33. AI AND ANDROID ~80% right. My experience of small, clear tasks. Not a benchmark. 33
34. REALITY CATCHES UP Then it got complicated. Simple tasks were only the beginning. 34
35. REALITY CATCHES UP Face the right way. One job. 35
36. REALITY CATCHES UP The comment isn't the behaviour. THE COMMENT “With g = -3000 px/s^2, choose v0 ≈ 620 px/s for peak height ~64 px (≈1.6 tiles),” var v = 1000f // px/s upward val g = -3000f // px/s^2 Read what it does. Not just what it says. 11a1962 · MainActivity.kt · verbatim comment line and code 36
37. REALITY CATCHES UP Describe the behaviour you want. var v = 1000f ↓ val targetJumpHeightPx = unitPx * 4f var v = sqrt(2f * -g * targetJumpHeightPx) A more explicit target — with a different intended jump height. 11a1962 · MainActivity.kt → 6f37070 · GameScreen.kt · inline comment omitted 37
38. REALITY CATCHES UP Connected isn't the same as shared. Two phones need the same meaning for “here”. BEFORE · MIXED UNITS fun sendPos( AFTER · GRID + FACING fun sendPos( xDp: Float, col: Int, row: Int, heightPx: Float facingRight: Boolean ) ) 29bfd5f versus its parent · MultiplayerManager.kt · signatures reflowed; bodies omitted 38
39. REALITY CATCHES UP Two tiny Chaos Monkeys. Unexpected players. Unexpected bugs. 39
40. REALITY CATCHES UP Build with kids. Fix after bedtime. 40
41. REALITY CATCHES UP “Go to the shop and get some bread, and if they have eggs, get six.” They had eggs. So: six loaves. Ambiguous instructions. A very literal result. 41
42. REALITY CATCHES UP Review is still my job. Architecture. Boundaries. Testing. Judgement. 42
43. TRY AGAIN 06 Life happens. Then you try again. The pause is part of the story. 43
44. TRY AGAIN The tools changed. So did I. 01 02 03 Frameworks Agents Experience Another attempt A different workflow Fewer unknowns 44
45. TRY AGAIN Same game. Different device. Shared Kotlin + Compose World · game rules · rendering · UI Android iOS Entry point + transport Entry point + transport 45
46. TRY AGAIN Their game. On an iPad. The device they wanted to play on. 46
48. HOW WE WORK NOW One idea. A small attempt. Play it. Record → choose → build → check → play Then decide what to change. 48
49. WHAT LETTING GO MEANT 07 It was always our game. We all had something to do. 49
50. WHAT LETTING GO MEANT The hardest part was avoiding coding. 50
51. WHAT LETTING GO MEANT Their game. Their friends. Their T-shirts. Leo wanted to support other indie developers. Free gameplay. Optional tips added for Shipaton. Released on Android and iOS. 51
52. WHAT LETTING GO MEANT There are more stupid ideas. Next: a workbench and a few recipes. Then see what they actually do with it. 52
53. TAKE IT WITH YOU Links, code and the slides. Don’t Go To Bed — Android + iOS jamescullimore.dev/dontgotobed/ Open-source code github.com/LethalMaus/DontGoToBed Get the game Elchem jamescullimore.dev/elchem/ Last year’s Droidcon talk Building a Mobile Game Without an Engine, a Budget, or a Clue Android iPhone / iPad Watch the previous talk This deck will be available on Speaker Deck. 53
54. WHAT LETTING GO MEANT Start. Fail. Learn. Try again. Thank you James Cullimore 54