The shift isn’t just toward smaller, faster and cheaper models. If intelligence can run locally—in products, devices and individual process steps—the real transformation challenge becomes designing the interplay of people, models, agents and decisions.

MIT Technology Review@technologyreview.com · last mo.

From sparse attention and rolling-summary memory to diffusion-based text generation and worm-brain-inspired networks, a wave of companies is quietly rebuilding the engine underneath the AI boom — each claiming meaningful performance gains at a fraction of the cost.

A pattern I keep seeing in transformation: The strategy is often not the problem. The missing layer is translation — from strategy into decisions, ownership, work, dependencies and governance. That’s where transformation becomes executable. Or remains a presentation.

I’m here to think in public about a question that has shaped much of my work: How do we make complex transformation actually executable? Expect observations on transformation, programme delivery, organisations, leadership and AI — mostly from the messy space between strategy and reality.