2025 may go down as the year AI went from the frontier to our daily workflows. 

At Columbia Business School, faculty across disciplines have spent the past year grappling with that shift firsthand. To close out 2025, we asked them to step back and reflect on a year in which AI moved decisively from experimentation to everyday use.

Their reflections are marked by a sense of acceleration and constraint: surprise at how quickly AI has crossed technical and cognitive thresholds, paired with growing clarity about the bottlenecks—energy, integration, organizational readiness, and human judgment—that now define its real-world impact. 

This article is part of a series in which CBS faculty reflect on the past year in AI.

CBS: Which development in AI surprised you the most this year and how has your own adoption of AI changed? 

The contribution of AI data-center investment to U.S. economic growth. Although investment in information-processing equipment and software amounted to only 4% of U.S. GDP in the first half of 2025, economist Jason Furman noted that it accounted for almost 92% of U.S. real economic growth over that period and likely crowded out other forms of capital spending. 

Without that AI-driven investment, Furman argued that the US economy may have ground to a standstill in H1-2025. This finding looks less stark once one adjusts the national accounts data to take account of the imports of tech equipment that are required to execute on the AI boom. But even then, it’s striking that a digital component of the economy has become such a core driving force in the evolution of the U.S.’s tangible economic growth.

  • Brett House, Professor of Professional Practice in the Economics Division

What concerns me, and perhaps even surprised me, is the confusion between hope and hype. We are all bombarded with exciting and hopeful promise of AI’s benefits and transformative impact. But when it is more marketing and hockey stick charts, the hype factor emerges. The “cautions” are only now being exposed, that the ROI may be slower than forecast, that the failure rate is significantly high for many pilots, and that the pressure to “do” something is not matched with the attention to how to implement and bring people along. 

Professor Stephan Meier and I are focusing in our work on how to manage the change, how to model employee centricity in the process of adopting these changes, how to equip and train employees, and how to invest in not only the product but also the process of introducing and embedding AI.

  • Todd Jick, Senior Lecturer in the Management Division and Reuben Mark Faculty Director of Organizational Character and Leadership at the Bernstein Center for Leadership and Ethics

This has been the year for foundational developments in organization and strategies due to leap-frogging technical change in AI. Agentic bots powered by large language models have penetrated business, including educational services offered at CBS. AI is entering a wide variety of markets faster than what computers did 40 years ago. 

AI has greatly amplified the automation of highly skilled and innovative tasks done by highly intelligent agentic bots. We have scrambled at CBS, faculty and students, to figure out the implications for how and what we teach to prepare our students and executives for the future of work.

  • Bruce Kogut, Sanford C. Bernstein & Co. Professor of Leadership and Ethics in the Management Division

I have been bullish on Gen AI since it emerged. And it has begun to deliver to an extent in many business and scientific contexts, especially when paired with a human expert: For example, this summer, I was able to prove a theorem in optimal transport theory, thanks in part to a month long conversation with Gemini 2.5 Pro where I used it variously as a tutor, assistant, ideator and sparring buddy. 

I must say, I've been surprised by substantial limitations of Gen AI including a seemingly fundamental inability to learn new skills, and contrary to my belief a year back, I now think Artificial General Intelligence is not around the corner. Rather, we need to understand and work with the limitations of this powerful technology, and patiently and systematically unlock the value it can provide. 

  • Yash Kanoria, Merrill Lynch Professor of Workforce Transformation in the Decision, Risk, and Operations Division

I was most surprised by the coding agent Claude Code and related developments from other labs (Codex, Copilot, etc.). I have started using it extensively, not just for coding, but even so far as to translate non-coding tasks to coding-like so that I can leverage these agents.

  • Will Ma, Roderick H. Cushman Associate Professor of Business in the Decision, Risk, and Operations Division

The ability to make progress with image generative AI was the biggest surprise. Another one is Google's move to embed Gemini in Google Search, giving us the right answer instead of a list of possible sources to a question we ask. 

I am using AI more seamlessly in everything I do, from searching for information, to writing and polishing writing, to coding, data analysis, and research.

  • Oded Netzer, Vice Dean of Research and Arthur J. Samberg Professor of Business in the Marketing Division 

The most surprising development to me has been personalization, or the ability of AI to “humanize”, i.e. to create a human-like creature that talks and behaves just like a real human being. Since this is simply the first example, we must realize the immense future importance of this development, particularly if/when this human “clone" can be imbued with the cognitive and reasoning capacities that AI can already bring into play.

  • Willie Pietersen, Professor of Professional Practice in the Management Division

The most surprising development for me was seeing AI systems reach gold-medal-level performance on the International Olympiad in Informatics—an achievement that once required years of specialized training. I spent my entire teenage years working on these problems and know firsthand how difficult they are, so watching AI reach that level is astonishing.

My own adoption of AI this year has been driven largely by vibe-coding. I’ve brought many new AI tools into the classroom to support coding, exploration, and faster iteration in teaching.

  • Tianyi Peng, Assistant Professor of Business in the Decision, Risk, and Operations Division

The pace of improvement. GPT5 was so much better than GPT4. Gemini now is also much better than at the beginning of the year. At the beginning of the year, I had mentally dismissed AI as a fad. At the end of the year, I am blown away by what it can do, to the point of thinking that it will change everything in the world of education, finance and accounting. 

  • Shivaram Rajgopal, Chair of the Accounting Division and Roy Bernard Kester and T.W. Byrnes Professor of Accounting and Auditing

OpenAI seems to be losing its edge, in particular with Google’s launch of Gemini 3 and Anthropic’s steady gains in reputation. In terms of my own adoption, the greatest change was using GenAI to help me write Python code.

  • Olivier Toubia, Glaubinger Professor of Business in the Marketing Division