The task is no longer to compete with AI on routine work. The task is to direct it, critique it, and go one level above it.
Manifesto
Research and training in the AI phase transition.
This page sets out the working stance behind the lab: AI capability has changed the baseline, the meaning of doctoral training is shifting, and the durable edge now lies in agency, critical thinking, judgement, taste, and invention.
Agency
Critical Thinking
Judgement
Taste
Invention
Core claim
2026 is treated here as a pivot year.
It is better to become future-proof now than to optimise for skills and research habits that are already being automated away.
AI is baseline
AI coding, implementation, ideation, and many early-stage research moves should now be treated as baseline capability. They are not side tools; they are part of the working environment.
The PhD is changing
A PhD still requires novelty and substance, but now novelty must be judged against both the literature and what strong AI systems can already ideate and implement.
The durable edge is human
The most important remaining advantages are agency, critical thinking, judgement, taste, and the ability to direct systems toward work that matters over time.
What Changes
Research should move one level above routine capability.
The question is no longer whether AI can code, implement, or produce incremental improvements. The question is what remains after a strong human researcher and a team of AI agents have already explored the obvious space.
Treat AI output as the new floor
If AI can already generate a comparable idea or implementation, that is not a result to be proud of by itself. It is the new starting point.
Ask deeper questions
Move away from small benchmark gains and toward questions that still matter after the current model cycle, especially those with conceptual depth or long test-of-time potential.
Build with agents, not despite them
Use AI aggressively for coding, ideation, implementation, and exploration, but direct it with strong judgement and a long-term research plan.
- A PhD means becoming an independent investigator, not just a competent implementer.
- You should be able to conceive ideas, form a plan of attack, execute with AI and other tools, and sell the work as a paper, patent, or product.
- Good work should have shelf-life. If it will feel obsolete in a few months, it is probably not the right target.
- Learn meta-skills with long durability: invention, mathematics, philosophy, people skills, product judgement, and communication.
- Assume coding and implementation are already heavily AI-accelerated.
- Optimise for questions that stay nontrivial even after strong AI assistance.
- Use AI-generated solutions as baselines and move beyond them quickly.
- Focus on agency, critical thinking, judgement, taste, and direction-setting as the real scarce resources.
Practical View
Different futures call for different preparation.
Industry path
Practice high-end AI-native building now. Build large, messy, real systems that touch hardware, humans, and operational constraints.
Academic path
Project five years ahead. Stay at the frontier, choose deeper gaps, and avoid competing on present-day skills that are already flattening out.
Personal stance
Your most important asset is time. Use it boldly. Tools are ready. The task is to develop judgement, taste, and the skill of invention before the window closes.