The AI replacement story is attractive because the first experience now feels real. A CFO sees Claude summarize reports and pull signal out of email. A COO sees scattered operational updates become a daily digest. For the first time, the AI feels like a real coworker, not just a chatbot.
Once you see that, the spreadsheet practically writes itself. A junior analyst costs this much. A subscription costs that much. Why not replace some human work with cheaper AI work?
Some of that math will hold. This is not a story about AI failing. But the spreadsheet has a hidden assumption: AI stays cheap, available, and reliable as usage scales.
The AI Replacement Math Got Less Clean - The main essay. AI work is not magic labor. It is inference. Every long prompt, document upload, agent loop, retry, tool call, and generated answer consumes compute somewhere. OpenAI says it has passed its original 10GW Stargate commitment. Anthropic is committing more than $100 billion over ten years to AWS technologies. GitHub says Copilot's old premium request model is no longer sustainable as coding agents move into longer sessions. The point is simple: if AI labor is compute labor, then replacement depends on the price, availability, and reliability of compute.
Also in this issue:
- Signals This Week - Agent control planes became a real category. The AI cost story got less simple. The first request is often not the real problem.
- The Wire - Google Cloud named the agent control plane. DeepSeek made the cost question harder. The Economist put numbers around the AI supply chain crunch.
- What I'm Consuming - Claude Code token budgeting, The AI Daily Brief on the end of AI subsidies, and the FT on authorial AI.
- After Hours - A double feature: Six Days of the Condor, the 1974 James Grady novel, and Three Days of the Condor, the 1975 Sydney Pollack film.
The practical answer is not to stop using AI. It is to replace tasks before roles. Measure cost per useful output, not just subscription price. Ask how much human review is still required, and what happens when the model is throttled, down, repriced, or moved behind a higher tier.