There is one universal truth in surfing and in life: You cannot catch a wave by paddling after it. You have to be in front of the wave when it forms.
The waves others are riding look so perfect. It is natural to want to ride them, too! But you are too late. The best thing that you can do is to position yourself well and be ready for a new wave to form. That is to say: Riding a wave is always a mix of luck, the ability to read the water, and being ready to take advantage when that next wave forms.
Don't be Tempted To Go Where the Action Is Now
What does that mean for AI/OR folks?
- Running after LLMs right now is paddling after the wave. The next revolution will come from a different angle. Not from transformers. Not from deep learning. Not from scaling "laws." Read Gary Marcus' take on the leaked Claude code and you will see that a new wave is already forming.
- OR is both taken for granted and under-utilized. Partly the latter is the case because several industries have given the old OR MIP-tech stack a look and moved on because it was not up for the task. If you are new to OR, the most wasteful thing you can do right now is to learn to model in mixed integer programming:
- LLMs already do it well.
- Literally every OR consulting shop falls back to the exact same stack of MIP solver technology and offers zero differentiation.
- The applications for which MIP tech is useful (and those exist!) have been milked to death.
Reading the Water
What you should do instead:
- The next wave (not just in OR but in business!) is decision-making under uncertainty. That is the holy grail that people were after when they were hoping ML could provide perfect 100% accurate forecasts. And they did not get it. But the need is still there, and it is massive. Every business has dozens of decisions to make every day without perfect information.
- Study stochastic optimization, the intricately connected multi-objective optimization, optimization under fixed time constraints, and sequential decision making (see the excellent books by Warren Powell and the "Decision Factories" book by Adam DeJans Jr. and John Brandon Elam).
- Learn to solve problems that no one else knows how to solve with legacy solvers. Take a look at our puzzles and you will see the wealth of real-world twists that you have been shielded from by textbook optimization formulations which miraculously are a perfect fit for MIP (alas not for any real application). You will find that there is new solver technology now that allows you to extend the convex hull of problems that can be solved with optimization.
The current LLM and MIP waves are ebbing. It is time to position yourself for the next waves.