Three months after OpenClaw hit 160,000 GitHub stars and proved that autonomous AI agents were not just a research project, the ripple effects are reaching Google's org chart.
Google is restructuring the team behind Project Mariner, its browser-based AI agent, as the company shifts resources toward coding agents. The move reflects a broader industry recalibration: the hottest AI product category right now is not chatbots or image generators, but agents that can actually do work on your computer.
The OpenClaw Effect
OpenClaw started as an Austrian engineer's hobby project. It connects large language models to real software through messaging apps like Signal, Telegram, and Discord. Unlike a chatbot that explains how to do something, OpenClaw carries out the steps - reading files, running commands, browsing websites, sending emails. By February 2026, it had become one of the fastest-growing open-source projects in history.
The creator, Peter Steinberger, announced on February 14 that he would be joining OpenAI, with OpenClaw moving to an open-source foundation. That single hire tells you everything about where the industry thinks the value is.
Every major AI lab has since accelerated its agent strategy. OpenAI's Operator reportedly achieves 87% success rates on complex browser tasks. Anthropic's Claude can control computers and orchestrate teams of sub-agents. Google's Project Mariner, which runs on Gemini 2.0, can handle up to 10 concurrent tasks on cloud-based virtual machines for $249.99/month AI Ultra subscribers.
Browser Agents vs. Coding Agents
The team restructuring at Google suggests a specific bet: coding agents may be more commercially viable than general browser agents, at least right now.
Browser agents face hard problems - websites change constantly, authentication flows are messy, and one wrong click can have real consequences. Coding agents operate in a more structured environment where mistakes are catchable through tests and version control.
The practical result for anyone using AI tools daily: expect coding-focused agents to improve fastest over the next year. Tools like Claude Code, Cursor, and GitHub Copilot's agent features are where the top talent and biggest R&D budgets are flowing. General-purpose "do anything on my computer" agents are still coming, but they are clearly the harder problem, and companies are prioritizing accordingly.
Google is not abandoning Project Mariner. But when a company reshuffles a team in response to a competitor's open-source project going viral, it signals that the internal roadmap just changed.