You can offload a task, or even a job, but you can never offload your learning

Nadella’s sharpest line cuts against the instinct to chase model selection: “the real opportunity is not in picking the best model but instead in building a learning loop on top of models where human capital and token capital compound. You can offload a task, or even a job, but you can never offload your learning.” A task or even an entire job can be handed to an AI, but the learning generated by doing the work is the asset that stays with the firm — “the future of the firm is the ability to compound that learning across people and AI.”

This reframes where to invest: not in the model layer (fungible) but in the retained loop. It is why The system of work is the moat, not the model — the model is fungible underneath and why Persistent agent memory preserves institutional knowledge that walks out the door with employees that would otherwise walk out the door with employees. The compounding nature of retained learning is the same mechanism behind Compound engineering makes each unit of work improve all future work and explains why The context flywheel is a Day 90 moat — Day 0 comparisons are misleading — the value accrues to whoever keeps the loop, not whoever rents the smartest model this quarter.