# Local Agentic Theory For Mobile Games — Shafik Quoraishee & Joanne Song, The New York Times

Source: [AI Engineer](https://www.youtube.com/watch?v=418t26CVz-w)  
Feed7 permalink: https://feed7.dev/p/local-agentic-theory-for-mobile-games-shafik-quoraishee-joanne-song-the-0s9z5h5  
Published: 2026-07-23T03:00:01.000Z  
Trust: Source Linked (source_linked)

## Why Included

Experimental on-device agents can play games and adapt interfaces without cloud calls, but real-time use must fit memory, frame-time, and battery budgets. Accessibility is promising, not production-ready.

## Source Summary

The prototypes run agents locally for Space Invaders and a mini crossword, with the latter using a constraint graph and backtracking. A proposed accessibility agent could detect interaction friction and adapt controls from live device context.

## Practical Implication

Design local loops around three hard budgets: model and state size, **16 ms per frame at 60 Hz**, and battery use. For accessibility, ground adaptations in **WCAG 2.2** and test predicted layout changes before applying them.

## Agent-Ready Context

The prototypes run agents locally for Space Invaders and a mini crossword, with the latter using a constraint graph and backtracking. A proposed accessibility agent could detect interaction friction and adapt controls from live device context.

Design local loops around three hard budgets: model and state size, **16 ms per frame at 60 Hz**, and battery use. For accessibility, ground adaptations in **WCAG 2.2** and test predicted layout changes before applying them.

This is experimental work and is not shipped in New York Times games. Faster planning, predictive models, long-term personal memory, a shared game-state language, better chips, and credible benchmarks are all still required.

## Context Map

- Layer: agent
- Domains: coding, image
- Topics: reasoning, computer-use, agent-reliability

## Uncertainty

- This is experimental work and is not shipped in New York Times games. Faster planning, predictive models, long-term personal memory, a shared game-state language, better chips, and credible benchmarks are all still required.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.