AI Business Radar — Tuesday, July 14, 2026
· CompaniesAutomation
Nadella warns that companies pay for AI twice—with money and the know-how they reveal—the "100x problem" of agents spikes bills despite falling token prices, and Anthropic extends Claude Code limits: the fine print of AI costs.
Yesterday's radar left the war for talent in the courts; Monday revolves around the fine print of the AI bill. Satya Nadella warns that companies pay for intelligence twice: with money and with the know-how they reveal to the provider. Meanwhile, token prices continue to fall — DeepSeek has already solidified its 75% cut — but agent bills are rising, and Anthropic is extending the expanded limits of Claude Code for another week because capacity still rules.
Nadella warns: you pay for AI twice — with money and with your know-how
Satya Nadella published an essay on X this Monday titled "The Reverse Information Paradox" in which he warns that companies "pay for intelligence twice: once with money, and once with something even more valuable, the proprietary knowledge they must reveal for that intelligence to be useful." He's not just talking about data: the prompts, workflows, evaluations, and corrections your team makes become institutional know-how that ends up in the hands of the model provider, and he argues that "what you create should belong to you." For your company: before connecting AI to your processes, check what the contract says about using your interactions to train models (and disable it where possible), and treat your library of prompts and flows as intellectual property — documented, versioned, and with controlled access. That this warning comes from the world's largest seller of corporate AI is, in itself, the news. Source
Tokens are getting cheaper, your bill isn't: the "100x problem" of agents
The 75% price drop that DeepSeek made permanent for its flagship model should be unqualified good news, but AI software providers are discovering the opposite: some vendors are reporting negative gross margins on their power users. The reason is that an agent converts a user query into a chain of planning, searching, tool usage, and verification that multiplies consumption — there are systems spending hundreds of times more tokens per query than their pricing model assumed, faster than rates are dropping. For your company: budget AI by task completed, not by token price, and measure the cost per result of each automation; and if you buy SaaS with "unlimited" AI at a fixed price per seat, expect that price to rise or have caps imposed — the provider is losing money on you. Source
Anthropic extends Claude Code expanded limits: capacity is still the faucet
Anthropic extended for the second time this Monday the promotion that expands weekly usage limits for Claude Code by 50%: it will last until July 19 and counts automatically for Pro, Max, Team, and Enterprise per-seat plans across CLI, IDE, desktop, and web. Behind this is the new computing capacity the company secured with SpaceX (over 300 megawatts and 220,000 GPUs), allowing it to give away usage while OpenAI pushes GPT-5.6 and ChatGPT Work. For your company: if your team develops with Claude Code, you have another week of extra margin without doing anything; the underlying lesson is that usage limits come and go based on the provider's available capacity — do not size a critical process based on a temporary promotion and always have a tested alternative model. Source
What to watch tomorrow?
Gemini 3.5 Pro continues to point toward July 17 with its 2-million-token context — still without official confirmation from Google —, the AI Act's digital Omnibus remains pending publication in the EU Official Journal, and the rollout of ChatGPT Work to Plus and Business plans should be completed these days.