AI Coding Agents: Claude Code, Cursor, Copilot, Codex — 2026-09-29

Top developments

Copilot CLI 1.0.89 can now ingest Claude Code rules files

On September 29, 2026, Microsoft shipped Copilot CLI v1.0.89, which can load rule files authored for Claude Code and use them directly as instructions. This is a notable interoperability move — teams that standardized their agent workflows on Claude Code's rules format can adopt Copilot CLI without rewriting project conventions. For developer desks, it signals GitHub is competing on compatibility rather than forcing lock-in.

Microsoft revamps the Copilot app, moving key features to usage-based billing

Microsoft has consolidated its consumer chatbot and enterprise AI tooling into a single refreshed Copilot app, with major functions shifting to consumption-based pricing, as reported by ZDNET Japan citing Ed Bott on September 28, 2026. The overhaul reflects Microsoft's broader effort to find a winning AI strategy — CNBC reported on September 25 that the update is explicitly framed as chasing Anthropic in coding capabilities.

Cursor reportedly at $4 billion annualized revenue

Fresh reporting says Cursor (Anysphere) has reached roughly $4 billion in annualized revenue, less than four years after founding — described as the fastest SaaS growth on record. This is up sharply from the $2 billion+ annualized figure reported earlier in 2026, and it comes as all major vendors have moved toward usage-based billing. It's a key data point in the consolidation of the AI coding market around a handful of leaders.

Enterprise adoption: coding agents drive token spending spikes

A September 28 report notes that coding agent breakthroughs in 2026 pushed enterprise AI token spending dramatically higher — with anecdotes of individuals moving from under $50 a year to $1,000 in a single day — forcing companies to tighten spend-management rules. This matters because it explains why nearly every vendor (Copilot, Claude Code, Codex, Cursor) has shifted from flat subscriptions to rate-carded consumption pricing.

Enterprise buyers compared on IP indemnity and 500-seat costs

MarkTechPost published (September 26, 2026) a comparison of GitHub Copilot, AWS Kiro, Cursor, and Cognition agents for enterprise rollout, covering IP indemnity, prompt storage policies, audit logs, and 500-seat costs. Procurement criteria — not just benchmarks — are becoming the deciding factor for large deployments.

Local view

• Japan: ZDNET Japan covered the Copilot app revamp with its consumption-based billing model, framing it as Microsoft's attempt to unify consumer and enterprise AI in a single app.
• Japan: Tech-Noisy highlighted the Copilot CLI's ability to read Claude Code rules files as the day's top AI news, underscoring Japanese developer interest in cross-tool interoperability.
• China: Zhihu's weekly model tracker (covering 2026/09/21–25) logged fresh agent-model updates: GPT-6 Sol and GPT-6 Luna (OpenAI), Claude Opus 5.5 (Anthropic) and Grok 4.7, feeding a fast-moving domestic comparison ecosystem for coding agents.

Context & numbers

• Cursor: ~$4B annualized revenue, per fresh reporting; previously $2B+ by February 2026.
• Adoption vs. trust: 87% of developers use or plan to use AI coding tools, yet trust in output remains low — a tension suppliers must fix for enterprise renewals.
• 84% of developers use or plan to use AI coding tools; 46% distrust the output.
• Claude Code leads developer satisfaction: 46% "most loved" vs. 19% for Cursor and 9% for Copilot.

On the radar

• Cross-tool standards: Copilot CLI's acceptance of Claude Code rules files may pressure other vendors to adopt compatible agent-configuration formats — watch Cursor and Codex changelogs next.
• Update cadence: Opus 5.5, GPT-6 Sol/Luna and Grok 4.7 shipped within the past week per Chinese trackers; benchmark leaderboards (Terminal-Bench, WebDev Arena) are likely to reshuffle soon.
• Spend governance: rising token spend anecdotes suggest a wave of enterprise AI-coding budget policies and FinOps tooling announcements in Q4.