Use AI aggressively
There is no point pretending AI coding and implementation do not exist. Use them well, manage them carefully, and treat them as force multipliers for serious work.
PhD Resources
PhD resources for the AI phase transition.
A compact guide for students thinking about doctoral study in a world where AI can already code, implement, and assist with substantial parts of research. The bar is moving upward.
2026 Pivot
The meaning of a PhD is changing.
Doctoral training now happens in a world of AI agents, recursive self-improvement efforts, and rapid shifts in what is considered routine work.
Read the full manifestoDo not optimise for obsolete gaps
Incremental model tweaks and small benchmark gains are no longer enough. Ask what remains after you and a team of strong AI agents have already explored the obvious space.
Protect long-term relevance
Learn skills with long shelf-life: invention, mathematics, philosophy, people skills, product judgement, and the ability to choose deep questions early.
- Linear algebra, probability, optimization, and causality
- AI: A Modern Approach
- The Elements of Statistical Learning
- Probabilistic machine learning and deep learning foundations
- Computer vision, NLP, reinforcement learning, and robotics texts
Novelty and substance
A thesis still has the same two criteria: novelty and substance. What has changed is that novelty should now be judged against both the literature and the strongest AI-generated baseline you can produce.
Judgement over routine skill
Programming skill alone is no longer enough to differentiate strong researchers. Judgement, taste, direction-setting, and the ability to critique and redirect AI outputs matter more.
Inventive capacity
The most durable advantage is the skill of invention: asking deeper questions, finding overlooked gaps, and building work that lasts longer than the current model cycle.