← back
Local Agentic Theory For Mobile Games — Shafik Quoraishee & Joanne Song, The New York Times
Takeaway
Useful mobile game agents must fit local memory and frame-time budgets while preserving offline and privacy benefits.
Summary
- The speakers describe experimental on-device game agents and clarify that New York Times puzzles remain human-created without deployed AI game features.
- Local inference can reduce network latency, preserve privacy, support offline play, and enable device-local personalization.
- The talk contrasts reinforcement learning that updates weights with agent loops that reason over game state and tools using in-context information.
- A Space Invaders prototype follows perceive, predict, decide, and act stages; mobile deployment must budget model memory, state history, rendering, and roughly 16-millisecond frames at 60 Hz.
on-device-aigamesagent-loops
Original description
An agent runs entirely on your phone, no cloud, and plays Space Invaders, perceiving the scene, predicting the aliens, and dodging bullets in a loop. Another solves the New York Times mini crossword with a constraint graph that backtracks when the fills stop fitting. These are experimental, and the Times is emphatic that its actual puzzles are made by people with no AI in the games. The hard part is the budget: a local agent has to plan its next move inside a single 16 millisecond frame, without draining the battery or starving the renderer, or you get jank. Shafik Quoraishee and Joanne Song use that to argue accessibility and difficulty should stop being separate toggles. Model the game as a continuous negotiation and they become two ends of one dial the agent tunes to you in real time: it watches eye gaze and shaky taps, resizes controls on the fly, and breaks a keyboard trap the moment it detects one, all on the device. Their closing bet is that the future of AI is not one giant centralized brain but billions of small local ones, each shaped by the person it runs for. Speaker info: https://x.com/squoraishee / shafik-quoraishee https://www.shafikquoraishee.com/ Timestamps: 0:00 - Disclaimers and the experimental frame 1:55 - A short history of AI in games 2:49 - Why run the AI on the device, not the cloud 4:56 - From reinforcement learning to agentic play 7:00 - Demo: an agent playing Space Invaders 8:19 - The on device budget: space, time, and energy 11:15 - Demo: solving the mini crossword by backtracking 12:33 - Accessibility: from toggles to graded dials 14:40 - The agent tuning the game to you in real time 16:28 - What is still needed, and billions of local brains