Google open-sources ARTEMIS, letting AI agents control a real phone end to end
dr_cintas · x · 2026-09-21
Google open-sourced ARTEMIS, a tool that lets AI agents control a real Android phone like a human: give it a plain-English task and it sees the screen, finds elements, taps, swipes, navigates, and verifies outcomes. It connects to coding agents (Claude Code, Codex, Cursor, etc.) over MCP.
- 99%+ task completion on AndroidWorld, 100+ multi-step tasks
- 3-5 seconds per step in Flash mode
- Element targeting fusing accessibility, OCR, and vision
- Screenshots, execution traces, and diagnostics on every run
- Python SDK for test pipelines
Example: tell Codex to build an APK, install it, log in with a test account, walk the checkout flow, and flag unexpected popups with screenshots — directly useful for QA, mobile testing, and debugging.
More from coding & agent
- If your LLM's tool-use trace shows it browsing Bluesky, shut that down immediately — venturetwins · 2026-09-21
- The real bottleneck is memory and compression, not context: dev argues compaction is a crutch — JoelMahon · 2026-09-21
- Jev Playground Launches as Author Teases System 1 + System 2 Agent Harness Guide — omarsar0 · 2026-09-21
- Where should the trust boundary for AI coding agents live? Dev argues merges need human sign-off — Fantastic-Sleep-3352 · 2026-09-21
- Asking Codex to Write a HUMANS.md So AI and Humans Understand Each Other — BLUECOW009 · 2026-09-21
- DAIR.AI Ships Jev Beginner Guide and Playground to Counter Misleading Demos — omarsar0 · 2026-09-21