One of my lectures, titled Becoming an AI Researcher after the book, will be dedicated entirely to PhD advice. In this lecture I will draw directly on the book, covering topics such as how to find the right research problem, designing experiments in AI, using large language models responsibly in your research, and thinking about the stages of the PhD from start to defense.

What draws me to summer schools like IAISS is the time it gives to attendants for in-depth discussions. A residential school keeps lecturers and students in the same hotel for the full week, so discussion continues over meals and breaks, not only in the lecture hall. I had a similar experience teaching at IIIA-CSIC’s Multi-Agent Reinforcement Learning course in Barcelona, based on my other textbook, Multi-Agent Reinforcement Learning: Foundations and Modern Approaches.

If you are a PhD student or early-career researcher and interested in attending the summer school, note that the early registration fee applies until August 23, 2026, just a few days from now. You can find all the details, including the full lecturer list, program, and registration, on the IAISS 2026 website.

Hope to see you there!

To get the discount, visit the official CUP page of the book and enter code BECAR2026 at the checkout.

Feel free to download and share the discount flyer below.

Becoming an AI Researcher Promo Flyer

1.02MB ∙ PDF file

Download

Download

We covered a lot of ground over the course of about an hour. On the research side, we discussed LLM-based agents and multi-agent systems, the shift toward a “web of agents” where personalized agents browse, transact, and negotiate on our behalf; whether multi-agent systems can produce genuine collective intelligence, and how to verify and constrain agents as they become more autonomous.

We also spent a good part of the conversation on guidance for new PhD researchers in AI, much of it drawing on themes from my upcoming book, Becoming an AI Researcher. That included how to develop research taste, why chasing publication speed is a bad substitute for finding problems that actually matter, and where to draw the line on using large AI models for tasks you haven’t yet learned to do yourself, so you don’t lose the learning process along the way.

If you are working on agents, multi-agent systems, or just starting out in AI research, I hope there’s something useful in here for you.

I am thrilled to share that the book is now available for pre-order! You can find it on the Cambridge University Press stores (e.g. US store, UK store, Canada store, and more locations available). The book is slated to arrive in stores in November 2026.

The book and pre-order were announced in my recent keynote talk at the 4th CCF Conference on Multi-Agent Systems (CCMAS 2026) on 27 June 2026 in Suzhou, China. The conference also included a dedicated session with students to discuss PhD guidance. We had some very encouraging discussions on questions such as how to find a good research problem, how to deal with slow progress and “failures”, how to manage the relationship with the supervisor, and how to prepare for a career in academia or industry. It was clear from our conversations that students want strong, practical mentorship to navigate today’s fast-paced AI landscape, and I hope that my new book will be a helpful guide to students.

Side note: special thanks to China Machine Press for bringing copies of the Chinese edition of the Multi-Agent Reinforcement Learning book. I’m sure the students at the session were very happy to receive the copies!

More updates to follow as we get closer to publication. If you’re interested in doing a PhD in AI, from choosing research problems to writing papers, surviving setbacks, and building a research career, consider subscribing to follow along.

The central image in the cover is shaped like a tree, drawn in a style inspired by artificial neural networks, which are the standard model used in modern AI.

The tree is a natural symbol for growth, and growth is at the heart of what this book is about. A PhD is a process during which you develop the ability to ask important questions and to carry out a program of research to discover answers to these questions. This includes skills such as critical judgement, effective communication, and resilience.

At the same time, the branching structure in the tree represents the many paths a student can take. During a PhD, you constantly face choices. Which problem should you work on? Which methods should you learn? When should you persist in the face of failure, and when should you change direction?

These choices create branches in the tree, some of which lead to better outcomes than others. In the cover design, that idea is reflected in the different endpoints of the branches: some branches end in larger circles than others, representing better outcomes. Not every decision has the same payoff.

That, in a way, is the point of the book. The advice provided in this book can help you navigate the branching structure more wisely and efficiently. It can help you avoid unproductive paths, recognise promising ones, and make choices that improve your chances of reaching a strong outcome.

I also like that the image connects the human side of research with the technical world of AI. The visual language comes from neural networks, but the meaning is about personal and intellectual development. In the end, the book is about people learning how to do good research in AI.

So, the cover is meant to capture both: AI as the field, and growth as the journey.

I look forward to sharing the final version once it is ready!

If you are interested in doing a PhD in AI—from choosing research problems to writing papers, surviving setbacks, and building a research career—consider subscribing to follow along.

Start with the problem, not the method.

This idea is the focus of Chapter 3 of the book, and it reflects a common turning point many PhD students experience during their research training.

A common trap in AI research

When students begin a PhD in AI, they are often excited by powerful new methods. Transformers, diffusion models, reinforcement learning algorithms, and many other techniques dominate conference headlines. It is tempting to start thinking:

What can I build with this method?

This often leads to a pattern of methods-driven research: tweaking architectures, adding modules, adjusting loss functions, and running experiments to see whether a new variant improves benchmark scores.

There is nothing inherently wrong with methodological innovation. Many important advances in AI come from new methods.

But when the method comes first, the research can easily drift into something less meaningful: a method searching for a problem to justify it.

Problem-driven research

A stronger approach is to begin with a clear, well-defined problem.

Instead of asking “What new method can I invent?”, the question becomes:

What problem needs to be solved?

Once the problem is clear, methods follow naturally.

Many landmark advances in AI emerged in this way. AlphaGo, for example, was not developed simply to propose a new reinforcement learning architecture. It addressed a specific challenge: how to achieve superhuman play in Go despite the game’s enormous search space, which makes brute-force search infeasible.

Similarly, early work on large language models was motivated by a deeper question in representation learning: whether a single model could learn general-purpose language representations from massive amounts of unlabeled text and then transfer them across many tasks.

In each case, the problem shaped the method, not the other way around.

Four questions that guide problem-driven research

In the book, I describe four core questions that help structure research around a problem:

1. What is the problem?
   Define it precisely.

2. Why does it matter?
   Who benefits if the problem is solved? What can it enable?

3. Why do current methods fall short?
   What limitations remain in the state of the art?

4. What is your contribution?
   How does your approach address the gap?

This structure forms the backbone of many successful research papers. A strong paper clearly answers each of these questions.

Of course, identifying meaningful problems is not always straightforward. In the chapter, I describe ways of discovering research directions. Good research problems rarely appear fully formed; they usually emerge gradually through a combination of vision and exploration.

A shift in mindset

Moving from methods-driven experimentation to problem-driven research is an important step in becoming an independent researcher.

It changes how you evaluate ideas. Instead of just asking “Is this method novel?”, you start asking “Does this solve an important problem?”

This shift also affects how you design experiments, how you write papers, and how you decide which directions are worth pursuing.

As Albert Einstein and Leopold Infeld wrote in The Evolution of Physics:

“The formulation of a problem is often more essential than its solution.”

The point is not that solutions matter less. Rather, a well-formulated problem gives direction and meaning to the entire research process.

Glimpse into the book

This chapter is one of the conceptual foundations of Becoming an AI Researcher. The book aims to guide readers through the full PhD journey in AI—from finding research problems and designing experiments to writing papers, navigating conferences, and completing the dissertation and defense.

The book is currently in production at Cambridge University Press and is expected to appear in Q4 2026.

In the coming months, I will continue sharing ideas and lessons from the book. If you are interested in the process of doing AI research and in building a research career, consider subscribing to follow along.

A packed room at AAAI 2026

Our event at AAAI 2026 — How to Do AI Research That Matters — was a great success!

We had a packed room, lively questions, and thoughtful conversations that continued well beyond the official session. It was energizing to see so many PhD students, postdocs, and faculty reflecting not just on what we build in AI, but why and how we do research in the first place.

Below are a couple of photos from the session. The room was too small, and a queue formed outside!

The book enters production

Following AAAI, I returned to the manuscript for a final round of editing. Over the past months, several colleagues generously took the time to read draft chapters and provide detailed feedback. I’ve now incorporated that feedback, clarified several sections, and refined examples throughout the book.

The book is now officially complete!

Cambridge University Press has now begun the production process, which includes professional copyediting, typesetting, and cover design. This stage transforms the manuscript into the finished book you’ll eventually hold in your hands (or read digitally).

Publication timeline

The book is scheduled to appear in Q4 2026.

Between now and then, I’ll continue sharing updates here — including glimpses into the production process, reflections on themes from the book, and announcements related to talks and events.

Thank you for following along on this journey! More soon.

Together with colleagues Profs. Reuth Mirsky, Manuela Veloso, and Matt Taylor, I’m excited to host the social event How to Do AI Research That Matters at AAAI 2026 on January 23. This event is designed to inspire and equip PhD students and early-career researchers with practical advice for thriving in AI research—advice that I've also compiled into my upcoming book, Becoming an AI Researcher: Practical Advice for Graduate Students, which I'm excited to announce for the first time at this event.

What to expect:

The event will open with an inspirational talk by Manuela Veloso on the history of AI research and what it means to do ambitious, impactful work. Reuth Mirsky and I will then share practical guidance on research best practices, navigating the publishing process, and dealing with setbacks. Finally, Matt Taylor will host a panel discussion featuring experts from academia, industry, and government, who will share their perspectives on impactful research and offer advice for PhD students.

If you’re attending AAAI in Singapore, I hope you’ll join us for what promises to be an inspiring and practical session. See you there!

To be published by Cambridge University Press in Q4 2026
Book homepage: https://phd-in-ai.com