Microsoft Cancels Claude Code Licenses — AI Agents Too Expensive to Scale
Microsoft is canceling most direct Claude Code licenses for internal use, moving engineers back to GitHub Copilot CLI, according to The Verge. The reversal comes six months after the company opened access to Claude Code, encouraging thousands of developers, PMs, and designers to experiment with the agentic coding tool. It got too popular — and too expensive.
Key Details
- Microsoft's Foundry deal with Anthropic ($5B investment, $30B Azure compute commitment) is not affected. Only internal Claude Code direct licenses are being cut.
- Uber CTO Praveen Neppalli Naga told The Information in April that the company had already burnt through its entire 2026 AI coding tools budget in just four months, after incentivizing adoption via internal leaderboards.
- Meta has a leaderboard called "Claudeonomics" tracking AI token usage per employee. Amazon pushes "toxenmaxx" — maximize tokens consumed.
- Goldman Sachs forecasts agentic AI could drive a 24-fold increase in token consumption by 2030, up to 120 quadrillion tokens per month.
- Gartner predicts inference cost for a 1-trillion-parameter LLM will drop ~90% by 2030 vs 2025, but total enterprise AI spend will still rise because agentic models require far more tokens per task, and AI providers won't fully pass cost savings to customers.
Nvidia VP Bryan Catanzaro summed it up: "For my team, the cost of compute is far beyond the costs of the employees."
The Bottom Line
The economics of AI agents are inverted: cheaper tokens don't mean cheaper bills. As Microsoft and Uber are discovering, token consumption outpaces price declines when agents are deployed at scale. If you're budgeting for AI tools in 2026, plan for costs to increase 3-5x over baseline — unless providers shift to fixed-price or usage-capped plans.
📖 Read the full source: HN AI Agents
👀 See Also
Claude adds inline interactive charts and diagrams to conversations
Claude now creates custom charts, diagrams, and visualizations directly within chat conversations, allowing users to tweak and modify visualizations as discussions develop. The feature is available in beta on all plan types and appears inline rather than in side panels.
The Need for Relational Governance in Multi-Agent AI Systems
Current governance frameworks focus on identity, permissions, and kill switches, but fail to address coordination between agents. Salesforce's research shows agent-to-agent interactions require purpose-built solutions, while studies reveal warmth outperforms dominance in negotiations.
Claude Opus 4.1 scores 17.75% on SWE-Bench Pro's private dataset, highlighting memorization vs. reasoning gap
Claude Opus 4.1 scored 80% on SWE-Bench Verified but dropped to 17.75% on SWE-Bench Pro's private dataset of 276 tasks from 18 proprietary startup codebases. Scale AI's analysis found models were navigating by memory rather than reasoning on familiar repositories.
OpenAI Training Costs Projected to Exceed Anthropic's by 4-5 Times Annually
According to confidential financials reported by the Wall Street Journal, OpenAI expects to spend 4-5 times more on training than Anthropic each year for the next five years. The expense scale is described as mind-boggling.