Google shut down Project Mariner on May 4th, 2026, ending a 17-month run for the agentic web-browsing experiment first revealed in December 2024. The project's landing page now reads: "Thank you for using Project Mariner. It was shut down on May 4th, 2026 and its technology voyaged to other Google products." The retirement lands two weeks before Google I/O opens on May 19th.
Mariner's capabilities have already migrated. Google has folded the underlying agent tech into Gemini Agent, which can archive emails or book hotels on a user's behalf, and into AI Mode, the agentic layer of its search product. A separate Chrome feature called "auto-browse," shown earlier this year, can run multi-step tasks like flight-price research, though Google has not confirmed it runs on Mariner internals.
Before the shutdown, Google had pushed Mariner to handle up to 10 tasks at a time, a meaningful jump from the single-task demos at launch. That parallelism is the technical bar agentic browsers are now competing on, and it is the capability Google appears to be carrying into its shipping products rather than keeping siloed in a research preview.
Key facts
- 01Google shut down Project Mariner on May 4th, 2026, roughly 17 months after unveiling it in December 2024.
- 02The project's tech now powers Gemini Agent, AI Mode, and a new Chrome 'auto-browse' feature.
- 03Mariner had been updated to run up to 10 tasks at a time before being retired.
- 04The shutdown lands two weeks before Google I/O kicks off on May 19th.
- 05Google is competing with agentic browsing tools from OpenAI, Perplexity, and OpenClaw.
Mariner debuted in December 2024 as Google's answer to the broader agentic-browsing wave. Since then, OpenAI, Perplexity, and OpenClaw have all shipped agents that drive a browser on the user's behalf, turning what was a research demo into a competitive product category. Google's decision to consolidate rather than maintain a separate Mariner brand is consistent with how the company has handled prior Labs experiments — ship the tech under the Gemini umbrella once it works.
“Project Mariner ran for roughly 17 months, from its December 2024 reveal to its May 4th, 2026 shutdown, with its agentic capabilities now living inside Gemini Agent, AI Mode, and Chrome.”— Jaeden Schafer
The timing matters. I/O 2026 begins May 19th, and Google typically uses the keynote to announce new Gemini-branded surfaces. Clearing Mariner off the board removes a confusing parallel product line and gives Google a cleaner story: agentic browsing is a Gemini feature, not a separate experiment.
Auto-browse in Chrome is the most interesting downstream beneficiary. Browser-level agents have an advantage over standalone tools because they inherit the user's logged-in sessions, cookies, and payment methods. Chrome has roughly two-thirds of global browser share, which gives Google a distribution lever that OpenAI's ChatGPT Atlas and Perplexity's Comet do not have, regardless of model quality.
Gemini Agent, meanwhile, is being positioned as the consumer-facing agent product. The pitch — archive your inbox, book a hotel, complete a checkout — overlaps directly with what OpenAI's Operator and Perplexity's agentic Comet features promise. The competitive question is whether Google's tighter integration with Gmail, Calendar, and Search beats OpenAI's lead in raw model capability.
AI Mode is the third inheritance. Google has been steadily wiring agentic actions into its search experience, including, as we covered last week, pulling Reddit and forum content directly into AI Mode and AI Overviews. Mariner's task-execution layer plugs into that surface, letting search results trigger actions rather than just summarize pages.
What is unclear is how much of Mariner's stack actually shipped intact versus what was rewritten. Google did not respond to a request for comment on the technical handoff, and the company has historically described agent capabilities at a product level rather than a model level. The 10-tasks-at-a-time benchmark is the only public performance marker.
Skeptics will note that agentic browsing remains brittle. Across every vendor — Google, OpenAI, Perplexity, OpenClaw — agents still misclick, lose state on multi-page flows, and fumble authentication walls. Shutting down a research-branded product before that core reliability problem is solved means Google is putting the Gemini brand on whatever rough edges remain. There is no Mariner label left to absorb the failures.
For Google, the consolidation is a bet that distribution and integration beat purity of product. Mariner as a standalone surface was never going to outpace ChatGPT's user base, but Gemini Agent inside Android, Chrome, and Workspace reaches a different scale of audience. The May 19th keynote will reveal whether that bet comes with a model upgrade behind it or just a rebrand of the same agent.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
