Google terminates Project Mariner as agentic AI capabilities consolidate into Gemini Agent

Google has discontinued Project Mariner, an experimental artificial intelligence browser feature that was launched in December 2024. The project, which was designed to perform autonomous web tasks such as booking hotels and managing email inboxes, was officially terminated on 4 May 2026. A message on the Project Mariner landing page confirms the shutdown, stating that the technology has voyaged to other Google products rather than being removed entirely.

The closure represents a strategic consolidation of agentic capabilities within Google's existing ecosystem. The technology previously housed in Project Mariner has been folded into Gemini Agent and AI Mode. These integrated tools now possess the ability to execute multi-step actions, such as archiving emails on a user's behalf or assisting with hotel reservations, effectively bringing the experimental functionality into the mainstream product suite.

Earlier in 2026, Google demonstrated a similar capability called auto-browse within Chrome, which allows for multi-step tasks like researching flight costs. While the company has not confirmed whether this feature is directly powered by the underlying technology of Project Mariner, the timing suggests a strong link. This development appears to be part of a broader response to competitive agentic web browsing tools offered by rivals including OpenAI, Perplexity, and OpenClaw.

The decision to shut down the standalone project coincides with preparations for Google's annual I/O developer conference, scheduled to begin on 19 May 2026. With the removal of Project Mariner, Google may be making room for other AI features that could make their debut during the event. The company has not yet responded to requests for comment regarding the specific decision-making process behind the shutdown or the extent to which specific capabilities were retained versus modified during the integration.

Google had previously announced an update allowing Project Mariner to perform up to 10 tasks simultaneously, highlighting the ambition of the initial experiment. However, the move to integrate these functions into Gemini Agent and AI Mode signals a preference for unified platforms over separate experimental tools. This approach aims to streamline the user experience while maintaining the advanced agentic capabilities that were originally tested in the browser.