Antigravity Enterprise and CodeMender Put Google on a Direct Path Against Claude Code and Cursor

Google advanced yesterday (19) into the corporate agent orchestration layer with the launch of Antigravity 2.0 and the integration of CodeMender into the Agent Platform. This move enters an already mature revenue market: Claude Code leads with over $2.5 billion in annualised revenue, Cursor surpassed $2 billion in ARR with over one million paid users, GitHub Copilot boasts 4.7 million paid subscribers and is growing by 75% year on year, and Codex has exceeded the $1 billion mark. In corporate adoption, Claude Code and Cursor are tied at 18% workforce penetration in the January JetBrains survey, with marked differentiation by company size: Claude Code dominates small companies with 75%, while GitHub Copilot still leads the segment of over ten thousand employees with 56%.

It is against this backdrop that Google positions Antigravity. The clear differentiation lies in governance and native integration with Google Cloud. Antigravity 2.0 arrives with a dedicated desktop application, a lightweight CLI, and access via standard Google Cloud credentials. It centralises management, customisation, and orchestration of agents in a single console, and is now available within the Agent Platform with Google Cloud's data privacy protections applied by default. The formal entry into Gemini Enterprise is expected in the coming months.

The names mentioned by Google as adopters provide a sense of the bet being placed. Accenture uses Antigravity to abstract infrastructure complexity and automate deliveries. AirAsia Next reports that over 50% of production-ready code is already generated by agentic workflows on the platform. Deloitte has built autonomous and governed software engineering pipelines on the same foundation. PwC Advisory runs background pipelines for customised client solutions, and Monks described the transition as a move from manual coding to high-level orchestration. The composition of these cases is not accidental: Google chose to showcase adoption in Tier 1 consultancies, precisely the firms that have historically represented the entry channel for Anthropic and GitHub in large accounts.

CodeMender Aims at Vulnerable Code

The most unexpected move was CodeMender, a security agent originally developed by Google DeepMind and now integrated into the Agent Platform. The agent autonomously identifies vulnerabilities, recommends precise fixes, tests changes in an isolated environment, and applies patches to dependent systems upon human approval. It is currently being tested with Gemini Enterprise clients, with expanded availability expected in the coming months.

The competitive reading is unsettling for the application security community. CodeMender enters as a direct competitor to Snyk, Veracode, and GitHub Advanced Security, with the edge of agentic execution from start to finish rather than mere detection and suggestion. The difference is not superficial. In a traditional stack, the scanner identifies, the developer evaluates and applies, and the cycle repeats. CodeMender compresses these three steps into one, with human involvement only at the approval stage. For CISOs with a backlog of remediation, the savings are real. For consultancies selling hours of vulnerability remediation, the competition will shift to added value over the outcome from the agent, rather than the task of identification itself.

Pressure on the Middle Layer

The combination of Antigravity, CodeMender, and Managed Agents API redefines what Google understands as a corporate AI platform. Instead of selling models and leaving integration to third parties, Google packages the execution layer, the isolated testing environment via Managed Agents API, and specialised agents into a single console surface. The advantage for CIOs is reduced contractual friction: a single contract, single billing, unified DLP policy. The complication for integrators is that part of the work that historically justified margins will come packaged in the product, shifting the commercial discussion from implementation hours to operational and governance agreements regarding agents in production.

The final reading is that 2026 will not be the year a single coding agent emerges as the winner, but the year in which the major clouds commoditise the coding agent within the corporate AI bundle. Those who currently sell pure agents will need to decide whether to become a platform or accept the role of a white label within the orchestration layer of Google, Microsoft, or AWS.