use case · engineering
Rank'em Bot
Six seats that ship a mobile game. Ads, code, bugs, creatives.
Ryan Perry runs Rank'em with a team of six Grok Bots (and growing). Each has its own job, its own computer, and its own overnight todo. The game just passed 1,000 downloads across iOS and Android in a week.
The 15% nobody needs a bot for is the game itself. He started at Sensor Tower and Lion Studios (AppLovin) watching how mobile games actually get built: anyone can one-shot a Candy Crush clone with Claude Code. Then they're stuck with the other 85% — Meta Ads Manager and AppLovin, hundreds of ad variants a week, the board and progression, GKE and certs, Apple and Google review, and the crash that only happens on one device. He wrote those six seats as job descriptions and handed one to each bot.
Roster: Mobile Orchestrator is the manager — owns App Store and Google Play, routes work between the other five. Analytics Agent owns paid acquisition and is the only bot allowed to declare a finding; it never writes creative and never touches app code. Creatives Agent turns findings into ad specs and renders variants; it never buys the media. Rank'em Engineer owns the app and backend and takes a finding as a spec, not a suggestion. GCS handles GKE, certs, deploys, and rollbacks — nothing reaches players without going through it. Bug fix Agent sweeps Sentry overnight, fixes what is obvious, and escalates what is not.
What broke first was the analytics loop, so that was the first bot. He could not use the Meta Ads API because of verification issues. Rather than wait on Facebook support, he recorded himself uploading a creative through the Meta web UI once. The bot kept the click-path and was uploading and testing at least one new creative per day. Routines: 7:10 PM CT spend/installs/CPI recap across Meta and AppLovin; 6:30 AM CT PostHog room-size and Partner Challenge funnel, flagging when the ad and the app describe different games; Mondays at 9:00 a seven- and thirty-day CPI rollup against Apple and Google Play revenue. Winners and losers go to Creatives. Product findings go to Rank'em Engineer as specs.
Approvals stay on him. Analytics checks with him before anything that spends money. Creatives never buys media. GCS is the only path to players.
The same feedback loop produced both wins. The bots read that people were playing as couples, not large groups, swapped the party ad for the partner-cut, and took installs from $15 to $1 — 15x cheaper. They then shipped a bonus hint feature after proving higher scores correlated with better retention, and D7 retention moved about 4x. Next empty seats: a revenue bot, more ad networks.
Official guide on x.ai, Aug 25, 2026.



