Quanta Bits

This week’s episode looks at why the AI replacement math is getting more complicated.AI tools are starting to feel less like chatbots and more like junior analysts, but the economics are not as simple as “AI is cheaper than people.” Compute capacity, token costs, model availability, and human review time all matter.I also cover OpenAI and Anthropic’s massive compute commitments, GitHub’s move toward usage-based Copilot pricing, Google Cloud’s agent control plane, and why AI adoption may start to look a lot like cloud adoption: useful, fast-growing, and eventually a CFO problem.Plus: a quick After Hours detour into Six Days of the Condor and Three Days of the Condor.

Everyone says "AI adoption" but nobody means the same thing. This week: why the definitions problem is killing more AI projects than the technology, what BCG's survey of thousands of CEOs reveals about the "pragmatist" majority, and why MIT says 95% of AI projects never make it past pilot. Plus: the fastest-growing open-source project on GitHub is an AI agent that got banned by its own infrastructure provider, and a 1968 novel that reads like an AI governance case study in 2026.

Anthropic published actual data on who AI is displacing, and it's not who most people expect. The most exposed group? People with graduate degrees earning 47% more than average. Computer, math, business, legal, and management roles are all above 90% exposed. But only 33% of that work is actually being done by AI today. That 61-point gap doesn't close on its own. It closes when leadership decides how fast to push, what to measure, and who's accountable.Also this week: Jack Dorsey cuts 40% of Block's workforce and Wall Street rewards him. AI models fail nuclear de-escalation simulations 95% of the time. Google's NotebookLM generates full cinematic video from your documents. And Burger King puts an AI coach in every employee's headset.Companion to Quanta Bits Issue 6. Full newsletter with sources: quantabits.beehiiv.com