What is changing across AI coding agents?
Engineers credit Antigravity with carrying multi-step tasks through and fault it for quotas and reliability. Across 123 public accounts the record leans negative, and Wipro and Cognizant have since put it into their own engineering.
Google changed Antigravity's quota rules four times in eleven months, so budget against today's rules and recheck them each quarter. Here is how usage is metered on Google AI plans and Gemini Enterprise seats, and what to budget per engineer.
Claude Code leads coding agent use at work, and frontier models now finish multi-hour tasks on their own. What your team can hand an agent comes down to how much output your reviewers can check and which permissions the agent runs with.
Also covers Claude Code, Codex, Cursor, GitHub Copilot, Devin, Devin Desktop (formerly Windsurf), Cline, Gemini CLI, Junie and Jules
