Tim O'Reilly – Why It's Better for Us to Think of AI as a Tool Than as a Worker (www.oreilly.com)

At NVIDIA’s GTC keynote Jensen Huang argued AI isn’t a “tool” but a new class of “workers” that use tools—citing examples like Perplexity agents booking trips, Cursor using VS Code to pair-program, and an imagined “AI chauffeur” in robotaxis—and implied a market far larger than traditional software. Tim O’Reilly pushes back: complex software has long functioned as “workers” (Amazon’s search, payments, logistics and dispatch pipeline is a clear example), and today’s AI, while more general and powerful thanks to statistical pattern-matching and large models, is an evolutionary leap rather than a categorical break. He stresses that many AI deployments still need humans to initiate, evaluate and supervise work, and full autonomy in domains like loan underwriting or end-to-end coding remains risky and limited. The distinction matters for valuations, product strategy and labor: treating AI as a worker pushes firms toward automation and displacement; treating it as a tool emphasizes augmentation and solving harder problems. Technically, current LLMs excel at generalization and speed but show mixed results in structured tasks, require human-in-the-loop safeguards, and often operate alongside—not instead of—existing systems. O’Reilly’s takeaway: celebrate AI’s productivity gains and democratization (“bicycle for the mind”), but don’t overstate its novelty or immediate capacity to replace broad swaths of software-driven work.