Tech • AI • Robotics
AI agents are coming: Google, OpenAI and Anthropic go on the offensive
9/10TL;DR
Major AI players shifted decisively in May toward autonomous agents, competing on infrastructure, developer platforms, and enterprise integration rather than standalone chatbots.
KEY POINTS
Open-source momentum with OpenCode
OpenCode, an open-source coding agent, gained traction with 160,000 GitHub stars, over 900 contributors, and about 7.5 million monthly developers. It supports multiple models instead of locking users into one ecosystem and emphasizes privacy by not storing code or context. Its rapid update cycle signals strong community-driven innovation as a neutral layer for agent-based development.
Anthropic scales compute via SpaceX deal
Anthropic expanded usage limits for Claude Code and its API after securing infrastructure from SpaceX, tapping into a data center exceeding 300 MW and 220,000 Nvidia GPUs. The move highlights a growing bottleneck in compute capacity and underscores that AI competition increasingly depends on infrastructure scale, not just model quality.
Claude evolves into enterprise workflow agent
Anthropic introduced integrations with tools like QuickBooks, PayPal, Canva, DocuSign, Google Workspace, and Microsoft 365, enabling Claude to act within business workflows. It also released financial service agents, signaling a push into regulated industries. Human validation remains required before actions like payments, reflecting cautious deployment in enterprise contexts.
Google launches managed agents in Gemini API
Google introduced managed agents that package execution environments, tools, and reasoning into a single API call. These agents run in sandboxed Linux systems capable of coding, browsing, and file handling autonomously. Though still in preview, the release shows Google shifting from model access to full agent execution environments.
Shift from Gemini CLI to Antigravity ecosystem
Google announced the transition from Gemini CLI to Antigravity CLI, with a key cutoff date of June 18, 2026 for many users. While enterprise access remains, the broader strategy is to unify fragmented tools under Antigravity, simplifying its developer ecosystem and consolidating agent capabilities.
Antigravity 2.0 targets full agent platform
Antigravity 2.0 expands into a comprehensive agent platform with a desktop hub, CLI tools, SDK, and API integration. It enables multi-agent orchestration, background task execution, and workflow connections across services like Firebase and Android. Early reactions have been mixed, but the ambition reflects Google’s push to control the full agent stack.
OpenAI advances Codex as enterprise agent
OpenAI Codex introduced features like Goal Mode for long-running tasks, remote computer use, and contextual screen input via Snapshots. With over 4 million weekly users and adoption by firms such as Cisco, Dell, Nvidia, and Datadog, Codex is evolving into a core enterprise tool for code review, testing, and automation.
Google Spark introduces persistent personal agent
Gemini Spark is positioned as a 24/7 personal agent operating via cloud-based virtual environments. Powered by Gemini 3.5 and integrated with Antigravity, it can manage tasks asynchronously across devices, even when offline. Planned integrations with third-party tools suggest a broad personal automation layer.
Convergence toward agent “operating systems”
Across the industry, Google, OpenAI, and Anthropic are building full-stack agent ecosystems combining goals, tools, memory, permissions, and execution environments. Google focuses on Antigravity and Spark, OpenAI on Codex multi-surface agents, and Anthropic on Claude-driven enterprise workflows, while OpenCode offers an open alternative.
CONCLUSION
The AI sector is rapidly transitioning from chatbot interfaces to integrated agent ecosystems, where control over infrastructure, execution environments, and workflows defines competitive advantage.
Explain this