In a new essay, veteran tech analyst Benedict Evans pushes back against the Silicon Valley assumption that AI will dynamically and spontaneously automate away most enterprise work. His core argument: most people aren't tool builders, and the hard part of software was never writing code — it was knowing that a problem existed and figuring out what the solution should look like.
Evans draws a spectrum from "institutionalized" software (SAP, Workday) to "improvised" substrates (Excel, email, shared folders). Tasks migrate between these poles over time as they grow in importance and scale. AI, he argues, doesn't change this dynamic — it creates new choices and moves the thresholds, but the fundamental cycle of bundling and unbundling continues.
The essay also addresses why enterprise AI adoption has disappointed so far. Every big company gave everyone Copilot or ChatGPT, yet usage remains uneven. Evans compares this to giving everyone a PC with Lotus 1-2-3 in 1983 or a web browser in 1997 — the technology was transformative, but not in the way anyone expected from just handing it out.
For the AI industry, the implication is clear: the real transformation won't come from making tools easier to build. It will come from the unimagined use cases that emerge when an entire organization learns to think differently about what software can do.
