📰 AI News: Big Companies Are Discovering the Hidden Cost of AI Agents 📰
📝 TL;DR 📝 Microsoft is reportedly pulling back on Claude Code licenses, Uber burned through its 2026 AI budget in just four months, and enterprise AI bills are rising even as token prices fall. The lesson is simple: AI can save time, but only if you understand what it actually costs to use. 🧠 Overview 🧠 A new wave of reporting shows that large companies are running into a surprising AI problem: usage is growing faster than budgets. Tools like Claude Code, Cursor, and Copilot can make teams more productive, but agentic AI uses far more compute than a normal chatbot because it reads files, writes code, tests, retries, and loops through many steps. 📜 The Announcement 📜 Microsoft is reportedly cutting back most internal Claude Code licenses by June 30 and shifting developers toward Copilot CLI, its own in-house coding tool. Uber also reportedly used its entire 2026 AI budget in just four months, largely because AI coding tools became widely adopted across engineering teams. This is not a sign that AI is useless, it is a sign that companies are finally seeing the real cost of heavy AI usage. ⚙️ How It Works ⚙️ • Token based pricing - Many AI tools charge based on usage, not just a flat monthly subscription. The more the AI reads, writes, thinks, tests, and retries, the more expensive it becomes. • Agents use more tokens - AI agents are not just answering one prompt. They often run long workflows across files, tools, commands, and multiple attempts. • Usage can explode - When companies encourage employees to use AI more, bills can rise faster than expected. Adoption is good, but unmanaged adoption gets expensive. • Cheaper tokens do not guarantee cheaper bills - Even if the price per token drops, total spending can still rise if teams use far more tokens overall. • Internal tools look more attractive - Microsoft shifting toward Copilot CLI makes sense because it owns the tool. Big companies want more control over cost, data, and workflow integration.