AI Coding Agents: Claude Code, Cursor, Copilot, Codex — 2026-09-27
Microsoft launched a major Copilot app revamp this week, adding Code and Autopilot capabilities aimed directly at Anthropic's Claude Code, moving to usage-based billing. Meanwhile, a new GitGuardian report warns that AI coding agents are leaking credentials on developer endpoints, and fresh statistics put Cursor at $4 billion in annualized revenue.
AI Coding Agents: Claude Code, Cursor, Copilot, Codex — 2026-09-27
Top developments
Microsoft revamps Copilot app to chase Anthropic — and moves to usage billing
On September 25, Microsoft announced an updated Copilot app combining coding (Code), Cowork, and Autopilot capabilities, and confirmed customers will pay by usage for these new applications rather than a flat subscription — commercial Microsoft 365 Copilot is currently $30 per user per month. CNBC frames the move as Microsoft "still searching for a winning AI strategy" and explicitly targeting Anthropic's momentum. The detail that customers pay per use in the new app was picked up widely, and Chinese media reported the announcement helped fuel a stock rally adding some 900 billion yuan in market cap in a day for the ~$25 trillion giant.
GitGuardian: AI coding agents are leaking credentials on laptops
A GitGuardian report published September 25 argues that Cursor, Claude Code, and Copilot leave credentials on developer endpoints that conventional repo and CI/CD secret scanners miss — a blind spot created by agent working directories, config files, and local logs. For teams scaling agents, this means security programs tuned to scan Git history don't cover the new attack surface. Expect endpoint-aware secret scanning to become a procurement requirement.
Cursor hits $4 billion annualized revenue
A statistics roundup updated six days ago puts Anysphere's Cursor at $4 billion in annualized revenue, calling it the fastest-growing SaaS company in tech industry history — up from the $2 billion+ run rate it reported by February 2026. For context, GitHub Copilot still leads on total users with roughly 20 million users and 4.7 million paid subscribers as of January 2026, and led JetBrains' January survey at 29% workplace use vs. Cursor's 18%.
Adoption high, trust low: 87% use or plan to use AI coding tools
SD Times reports that 87% of developers already use AI coding tools or plan to, but widespread usage "does not necessarily translate into confidence" — echoing other surveys showing only around 29% fully trust AI output. A companion analysis in Forbes Tech Council (Sept 25) cautions that scaling AI coding tools doesn't help if the bottleneck is elsewhere in the delivery pipeline.
New agent entrants shake up comparisons
TechRepublic published a comparison of five Claude Code alternatives — Codex, Google's Antigravity, Copilot, Cursor, and Kiro — ranking them on pricing, security, and team controls (Sept 24). Tech-Insider added a comparison of Meta's new "Muse Code" agent against Claude Code and Cursor (Sept 26), a signal that Meta is entering the coding-agent fray. Morphllm's re-verified leaderboard (Sept 22) has Codex + GPT-6 Astra leading Terminal-Bench 4.0 at 58.2%, with Claude Code + Fable 5.1 at 57.9% at twice the cost; Opus 5.5 and GPT-6 Sol shipped the same week.
Local view
- China (Sohu): Coverage of the Microsoft Copilot update emphasized the shift from subscriptions to usage-based billing (Cowork, Code, Autopilot) and the resulting ~900 billion yuan single-day market cap jump for Microsoft. A Zhihu model tracker (Sept 23) logged the week's agent-model releases: GPT-6 Sol/Luna, Claude Opus 5.5, and Grok 4.7 — the model layer beneath the coding agents is churning fast.
- China (developer blogs): EastonDev (Sept 24) published a solo-founder workflow splitting Codex, Claude Code, and Cursor duties across planning, implementation, review, and launch — a popular "multi-agent stack" pattern rather than picking one tool.
- Japan: AI-Heartland (Sept 26) tested Microsoft's new Managed Apps CLI/SDK via npm to examine what Copilot Code and Autopilot actually ship under the hood. AI-Rise (Sept 25) ran a Copilot vs Claude Code feature/pricing explainer for Japanese developers.
Context & numbers
- Claude Code pricing tiers: $20 Pro, $100 Max 5x, $200 Max 20x, plus per-seat Team plans; API billing can be cheaper for bursty use.
- GitHub Copilot tiers (approximate per comparison guides): Pro $10/mo, Pro+ $39/mo, Max $100/mo with 20,000 credits (~$200 of usage); Business $19/user/mo; Enterprise $39/user/mo — note these are from search snippets, verify on official pages before purchase decisions.
- Cursor: $4B annualized revenue; Copilot: ~20M users, 4.7M paid subscribers (Jan 2026); JetBrains Jan 2026 survey: Copilot 29% vs Cursor 18% workplace use.
- Benchmark leaders (Sept 22 re-verified): Codex + GPT-6 Astra 58.2% Terminal-Bench 4.0; Claude Code + Fable 5.1 57.9% at ~2x cost. WebDev Arena references a 1,762 Elo score in Sept comparisons.
On the radar
- Meta's Muse Code just entered third-party comparisons; watch for official availability and pricing details in the coming weeks.
- Watch for endpoint secret-scanning tooling (from GitGuardian and rivals) adding first-class support for Cursor/Claude Code/Copilot working directories.
- Microsoft's usage-based billing for the new Copilot apps could reshape enterprise M365 Copilot contracts — 30-dollar flat seat pricing may not survive long.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
