Google quietly shuts down Project Mariner, its AI agent that navigated Chrome and completed tasks on a user's behalf, after highlighting it onstage at I/O 2025
NEW: Google quietly shut down Project Mariner yesterday, the web-browsing AI agent it highlighted onstage last year at Google IO. I reported for WIRED, nearly 2 months ago, that Google had moved staffers off the Project Mariner team as it responded to OpenClaw-style agents. [image]
Context & Ripple Effects
Project Mariner began as a DeepMind prototype for controlling Chrome, then became part of Google's public agentic-AI messaging at I/O 2025. By March, Google had already moved some of its staff to higher-priority work.
The shutdown arrives while Google has separately been preparing Chrome agentic features and describing safeguards for them, suggesting Mariner was not the only vehicle for its browser-agent work.
First-order effects
- Project Mariner is no longer an active Google product effort, ending a distinct prototype focused on completing browser tasks for users.
- The earlier staffing moves are effectively confirmed as a reprioritization away from Mariner rather than a temporary team adjustment.
Second-order effects
- Google's browser-agent development is likely to consolidate around other Chrome and Gemini initiatives, where action-level security and user-alignment controls have already been discussed.
- Competitors building browser-operating agents gain a clearer opening to differentiate on execution and product availability, while Google avoids maintaining a separately branded agent effort.
Third-order effects
- The episode underscores that browser agents remain a difficult product category to operationalize: capability demonstrations can be separated from durable, broadly deployed products by reliability, security, and user-control requirements.
- If major platforms increasingly fold agent behavior into existing browsers and assistants rather than ship standalone agents, competition may shift toward control of the browser, identity, and safety layers that govern automated actions.
The trend: Agentic AI is moving from high-profile prototypes toward more tightly integrated, security-governed features inside established consumer platforms.