Google employees have been quietly flocking to an internal AI tool called Agent Smith that automates coding and other tasks , so many of them that the company had to throttle access to keep up with demand.
Agent Smith, named almost certainly after the relentless villain from The Matrix, has become one of Google’s most sought-after internal tools since launching earlier this year. Three people familiar with the tool confirmed its existence, and two said access had to be restricted after the volume of employees trying to use it overwhelmed capacity. When a corporate AI tool gets rationed, that tells you something about how hungry the workforce is for it.
What makes Smith distinct from a typical AI coding assistant is how it works. Built on top of Google’s existing agentic coding platform called Antigravity, it operates asynchronously , meaning it keeps working in the background even when an employee has closed their laptop and stepped away. Engineers can delegate tasks, walk away, and check back in later. Better yet, they can monitor progress and issue new instructions directly from their phones. That’s a meaningfully different workflow from the current generation of AI pair-programmers that require you to sit there and babysit every suggestion.
The asynchronous angle matters more than it might first appear. Most enterprise AI coding tools today are fundamentally reactive , they respond to prompts but require constant human input to move forward. Agent Smith flips that model, functioning more like a junior developer you’ve assigned a ticket to than a sophisticated autocomplete. For a company the size of Google, where engineering backlogs are enormous, the productivity math here is compelling.
Agent Smith didn’t emerge in a vacuum. Cofounder Sergey Brin, who has become notably more present at Google in recent years as the AI race has intensified, has been publicly emphasizing the role of AI agents inside the company. He has pointed to tools that can act independently on complex tasks as a key part of Google’s internal productivity strategy. Smith appears to be a real-world implementation of that vision, not just a prototype.
Google has also been increasingly explicit with employees that adopting AI tools is an expectation, not a choice. That cultural shift, which mirrors similar pressure at Microsoft and other major tech firms, creates a ready-made user base for anything the company rolls out internally. The popularity of Agent Smith suggests that when the tools are genuinely useful, employees don’t need to be pushed very hard.
The name Antigravity, the underlying platform Smith is built on, also deserves a mention. Google has been developing its internal agentic infrastructure for some time, and Agent Smith appears to represent a consumer-facing layer on top of that foundation. The fact that it can hook into various internal tools , not just write code in isolation , suggests a more deeply integrated system than a standalone assistant.
What This Signals for the Broader Market
Google keeping Agent Smith internal for now is a strategic choice, but it’s hard to imagine the company sitting on this indefinitely. The agentic AI coding space is heating up fast. Startups like Cognition (which builds Devin, the autonomous software engineer) and established players including GitHub Copilot’s next-generation features are all racing toward the same destination: AI that doesn’t just assist developers but actually ships work independently.
If Agent Smith is as capable as its internal adoption rate suggests, Google has a commercially viable product on its hands. The question is whether the company moves to productize it for Google Cloud enterprise customers, integrate it into Workspace, or keep it as an internal edge for its own engineering teams. Given how aggressively Google has been commercializing its Gemini infrastructure, a broader release seems more likely than not.
For enterprise software buyers and CTOs watching this space, the takeaway is practical: agentic coding tools that work asynchronously are graduating from interesting demos to genuinely useful infrastructure. The fact that Googlers are competing for access to one internally is a stronger signal than any product announcement. Watch for what Google does with Antigravity and Smith in the second half of 2026 , and watch whether OpenAI, Anthropic, or the DevTools startups get there first on the outside.
Also read: Australian startup Syenta raises $26 million and lands Pat Gelsinger to tackle the AI chip supply crisis • OpenAI faces a mirror test as users prompt ChatGPT to visualize the most average human life • The political grift has gone algorithmic as scammers abandon stolen identities for limitless AI-generated personas.