How I made Kanto Tactics with AI

Kanto Tactics is a free autobattler you play in the browser: you buy Pokémon in a shop, place them on a board, combine their types and watch them fight. It launched in early October 2026 and, in its first weeks, more than 7,000 players reached the campaign ranking.

I made it on my own, with AI as my creative partner. This post explains how that works day to day.

One assistant per topic, and a director

I don't use one big chat for everything. Each part of the game has its own assistant with its own notes: balance, the Battle Tower, adventures, online PvP, mobile, the Discord community, security… A director assistant coordinates them, merges their work and prepares each release.

Every assistant leaves a handover note when it finishes, so the next one picks up exactly where the last one left off. That is what lets a single person keep a dozen fronts moving at once.

What I still decide

• The vision: what the game should feel like and who it is for.
• Every release: nothing goes live without my OK.
• Taste: art, feel, and when something is fun or just correct.
• The community: I read the feedback and choose what matters.

Tests instead of trust

AI writes a lot of code quickly, so the safety net matters more than ever. The battle engine runs hundreds of recorded games after every change and must give identical results, or the change does not ship. Online battles depend on it: every player's game has to simulate exactly the same fight.

Shipping fast

In the first weeks a new version came out almost every day: friend rooms, ranked seasons, the Battle Tower, adventures and many balance fixes. You can see all of it in the changelog.

If you want to see the result, play Kanto Tactics. If you want to follow how it is made, I share it on X and YouTube.