Deep dive sessions

AI Fluency What kind of AI fluency do future business leaders actually need?

The Case Method How should classic business education tools—like the case method—evolve when students can use generative AI to summarize and shortcut the process?

A New Role How should business schools prepare students for a world where AI is increasingly part of the work itself?

Product & Entrepreneurship As AI dramatically reduces the cost of building products and launching companies, what skills become more valuable?

Learn to Code — Or Don't If AI can build websites, write code, analyze data, and generate business applications, what skills should MBA programs prioritize—and what skills remains uniquely human?

Learn Differently How can educators design classrooms that make students’ use of AI more visible, structured, and intellectually rigorous?

Building AI fluency

The big question

What kind of AI fluency do future business leaders actually need?

MBAs may not have to become full-time AI engineers, but they do need more than chatbot familiarity or prompt-writing skills. They need enough technical intuition to understand AI systems and enough judgment to decide when and how those systems should be used.

Why it matters

AI has become part of the operating DNA of business. It is already reshaping products, analytics, marketing, finance, entrepreneurship, operations, and decision-making.

Executives and entrepreneurs increasingly manage teams, tools, and workflows built around AI systems they have not personally engineered.

That creates a new leadership challenge. To lead effectively, execs and entrepreneurs will need to ask sharper questions, evaluate AI outputs critically, and understand tradeoffs around cost and reliability.

Top takeaways

MBAs need to go ‘one level deeper’

Technical training for MBAs can’t be surface level, but that doesn’t mean every successful MBA will have to be an engineer, either. What they need is fluency. An approach that’s “light on math” and heavy on teaching students to develop and trust their intuition helps those students code models, test ideas, and make the mistakes they need to make to learn.
AI is much more than chatbots

Interfaces like ChatGPT, Gemini, and Claude offer a limited view of what AI can do. Tools like XLKitLearn, developed at CBS, give students a more multifaceted understanding by bringing AI capabilities into Excel. There, students can learn in a more hands-on way by comparing model responses, examining prompt costs, testing guardrails, using embeddings, and seeing how agents use tools to complete tasks.
Students need to be builders and evaluators of AI systems, not just users

The CBS MBA Copilot lets students build a retrieval-augmented AI assistant, using course materials and other documents. That exposes them to the infrastructure behind AI applications. In the same vein, the School’s AI List helps students discover, review, and organize tools by use case, helping them understand the AI ecosystem critically and customize systems for specific business needs.

The bottom line

Leaders must be prepared to pilot organizations where AI is deeply and widely embedded. That requires technical intuition, hands-on practice, and the judgment to know when an AI system is reliable, cost-effective, and aligned with the problem at hand.

What’s next?

Organizations will need to create more opportunities for leaders to experiment with AI systems and apply them to business problems across functions. As the technology evolves, they will need to move beyond just tool-specific training.

Using voice AI to reinvent the case study

The big question

How should classic business education tools—like the case method—evolve when students can simply use generative AI to summarize and shortcut the process?

Don’t abandon the case method altogether. AI can be used to make case preparation more active, more conversational, and more demanding.

Rather than giving students easier access to answers, use AI-enabled tools to create productive friction: ask students to take a position, defend their reasoning, and practice judgment before they enter the classroom.

Why it matters

The case method depends on preparation. Students read, reflect, form an argument, and arrive ready to debate. Generative AI disrupts that process by making it easy to bypass the struggle that often produces learning.

Voice-based AI discussion partners and interactive AI cases can encourage students to engage more deeply with ambiguity, disagreement, and decision-making. For instructors, these tools can also reveal how students think before class, making live discussion richer and more targeted.

Top takeaways

AI can make case preparation more active

CAiSEY, an AI-based conversation partner developed at Columbia Business School, immerses students in real-time discussion around a case question. Instead of submitting a written response, students debate an AI partner that challenges their argument and pushes them to consider another side. The goal is to deepen critical thinking before class, so students arrive having already practiced the kind of reasoning the case method is designed to teach.
Voice matters

CAiSEY’s voice-first format is central to the learning experience. Speaking with AI can feel more immediate and higher stakes than typing into a chatbot, so using CAiSEY encourages students to think on their feet. It can make participation more inclusive for students who speak English as a second language or who have learning differences that make writing challenging. Research on CAiSEY also suggests voice-based conversations can produce more varied ideas than text-based exchanges.
AI cases can turn static PDFs into simulations

Another way to rethink the case method is to replace or supplement static case PDFs with interactive AI cases. In this format, students speak with AI-generated characters inside an organization, such as a CEO, CFO, principal, or frontline worker. Students must ask the right questions, interpret conflicting perspectives, gather data, and make recommendations. The result is a more open-ended experience that can better reflect the messiness of real-world decision-making.

The bottom line

Used thoughtfully, GenAI can restore some of the valuable elements of struggle, debate, and preparation to the classroom experience. The strongest applications require students to reason, speak, listen, adapt, and defend a point of view.

What’s next?

The next step is designing AI-enabled case experiences around clear learning goals. Instructors will need to decide when students should debate an AI partner, when they should explore an interactive simulation, and how those interactions should inform classroom discussion. As these tools develop, the case method may become less static, more personalized, and more closely connected to the ambiguity of real managerial decisions.

AI-native startups: A course built around agent CXOs

The big question

How should business schools prepare students for a world where AI is increasingly part of the work itself?

Tomorrow’s leaders will need to know how to work with AI agents, evaluate outputs, and make decisions when information is incomplete or constantly changing.

Why it matters

AI is lowering the cost of building products, expanding what small teams can accomplish, and changing how work is delegated. Startups can now prototype faster, while founders can rely on specialized AI agents to increase productivity, and students can simulate workplace situations that once required far more time, people, or infrastructure.

Top takeaways

Students need to practice with AI agents, not simply use AI tools

AI-native startups may increasingly rely on specialized agents that can act like CFOs or CEOs. In a course model proposed by Professor Song Ma, students learn the core knowledge needed to design, test, and improve those agents, then apply them in simulations such as startup fundraising and investor negotiations. Students still need foundational knowledge, but the goal shifts toward building competent agents and knowing how to use them in real decisions.
AI can make business education more dynamic and realistic

Professor Kyle Maclean’s Hopeworks Foundation case uses AI to turn a static assignment into an iterative consulting-style experience. Instead of receiving all the information upfront, students begin with partial data, ask questions, receive new variables, run analyses, and revise their thinking. The format better mirrors the workplace, where leaders rarely get a complete packet of information before deciding what matters.
Entrepreneurship education has to move from planning to building

Professor Mattan Griffel emphasized that AI has dramatically lowered the cost and time required to launch a startup. Students can now build working prototypes in minutes, which makes the traditional business plan even less useful. The challenge is to teach students when to use AI for speed, when to use it as a tutor, and when to treat it as a thought partner.
The agentic workforce could make everyone a manager

AI agents turn individual contributors into managers of workstreams, according to Professor Paul Canetti. Entry-level employees may soon delegate tasks to AI systems, review outputs, and coordinate work performed by both people and agents. That means students need to practice scoping assignments, giving instructions, and evaluating results.

The bottom line

AI pushes business education toward more active forms of learning. Students will still need fundamentals. They also need repeated practice making decisions with AI in the loop.

What’s next?

More simulations, more agent-based assignments, more live building, and more opportunities for students to practice judgment under uncertainty.

Rapid prototyping with AI

The big question

As AI dramatically reduces the cost of building products and launching companies, what skills become more valuable?

Why it matters

AI compresses the traditional product-development cycle. Tasks that once required specialized technical expertise can now be completed by small teams—or even individuals—with little coding experience. As building gets easier, identifying the right problems and connecting with the right people becomes both more difficult and more important.

Top takeaways

Technology is no longer the hard part

LaSala argued that AI tools are shrinking every stage of the product-development process, from research and prototyping to testing and iteration. As a result, someone with no formal product or computer-science training can build apps, websites, platforms, and products with relative ease. The hard part is validating demand and finding customers.
Human feedback remains irreplaceable

Sturt described an experiment in "vibe entrepreneurship" in which MBA students used LLMs to launch real startups. Teams that secured conversations with potential customers refined their ideas, found product-market fit, and accelerated development. Teams that relied primarily on AI-generated customer simulations, synthetic data, or automated outreach often stalled. The biggest differentiator was access to real people who could provide honest feedback.
Product management shifts from execution to judgment

LaSala argued that future product managers will spend less time writing specifications and overseeing development and more time framing problems, evaluating tradeoffs, and ensuring teams are building the right thing. AI can help generate solutions, but it cannot determine whether a problem is worth solving.

The bottom line

AI makes entrepreneurship and product development more accessible than ever. But as the cost of building approaches zero, the scarce resources become customer insight, human relationships, strategic judgment, and the ability to identify meaningful problems. In this world, the competitive advantage may belong to those who can learn fastest from others.

What’s next?

The next step is designing entrepreneurship and product courses that treat fast building as the starting point, rather than the end goal. Students will need more practice testing ideas with real customers, interpreting feedback, and deciding when to pivot, persist, or stop. As AI makes prototypes easier to create, business schools will need to emphasize the harder skills: problem selection, customer discovery, judgment, and learning from the market.

Teaching MBAs to build with AI from day one

The big question

If AI can build websites, write code, analyze data, and generate business applications with little technical expertise, what skills should MBA programs prioritize—and what remains uniquely human?

Why it matters

AI is changing not only what students can do, but how they learn. Tasks that once required programming expertise can now be completed through natural-language prompting. That raises a fundamental challenge for business schools: determining which skills should still be taught directly, which can be augmented by AI, and which become even more important in an AI-native world.

Top takeaways

Building beats lecturing

Boussioux has increasingly shifted his courses away from theory-first instruction toward hands-on creation. Students build games, websites, AI agents, and analytics tools from the first sessions, with fundamentals introduced later, once students have experienced the technology’s capabilities firsthand. His goal is to make AI feel empowering by helping students turn ideas into working products.
AI dramatically expands what’s possible

Both speakers argued that AI allows students to tackle projects that would previously have been out of reach. Shumsky’s students build predictive models and decision-support tools without prior coding experience, while Boussioux’s students create websites and applications for real businesses and personal projects. But both emphasized that results depend heavily on how users frame problems and prompts.
Human judgment remains the bottleneck

Shumsky demonstrated that AI can produce convincing but flawed outputs by mishandling data, such as by removing meaningful outliers. Users must understand the underlying business problem, evaluate results critically, and apply domain knowledge that AI may lack. In analytics, knowing what question to ask may matter more than knowing how to code.

The bottom line

AI is lowering the barriers to building and analyzing, but as the technology grows more capable, the value of human judgment, problem formulation, creativity, and domain expertise is becoming more important than ever.

What’s next?

Students will need more opportunities to create working products and analytics tools, then pressure-test the outputs against real business problems, data limitations, and domain knowledge.

Using Loveable.dev in teaching

The big question

How can educators design classrooms that make students’ use of AI more visible, structured, and intellectually rigorous?

Why it matters

AI has moved much of student learning into private, unobservable spaces. Students may ask chatbots dozens of questions about an assignment, but instructors often see only the final answer. That creates what Beasley calls a “data opportunity cost”, meaning the loss of visibility into where students are struggling, what they misunderstand, and how they reason through material.

Top takeaways

Dialogue can become the assignment

Instructors can use AI tools like Socrat, developed by Luke Beasley, to create dialogue-based assignments. This allows instructors to set the stages students must work through and the criteria for advancing. The result is a more interactive form of assessment, where students demonstrate understanding through ongoing communication.
AI can give instructors a new feedback loop

Because Socrat keeps student conversations within a single ecosystem, instructors can analyze where students succeed or fall short. Beasley framed that data layer as one of the tool’s biggest advantages: It helps instructors see, in real time, which concepts need more attention.
Students need to learn how to question the machine

For Lee, the teaching challenge is not preventing students from pasting problems into LLMs but helping them become more discerning users. In his AI Foundations course, students use LLMs to analyze point-of-sale data for inventory planning, then examine the choices the models make. He emphasized that analytics work often does not have a single correct answer. The deeper learning comes from understanding what the model did, what assumptions it made, and how those choices affect downstream business decisions.

The bottom line

AI can make learning more personalized, conversational, and data-rich but only if educators redesign assignments around real-world applications of the tools. The best teachers don’t merely add AI to the classroom but teach students to improve the process of interacting with it.

What’s next?

Educators should build assignments that capture more of the learning process, not only the final submission. Design AI-enabled exercises that make students’ questions, reasoning, and misconceptions visible while also teaching them how to interrogate model outputs.

GenAI-driven oral assessments

The big question

If generative AI has made written assignments less reliable as evidence of student learning, can educators use AI itself to create better assessments?

Why it matters

Students can submit polished, comprehensive assignments generated with help from LLMs, and still be unable to participate meaningfully in class discussion.

That creates a problem for courses built around judgment, product management, or any other sort of open-ended decision-making. Instructors need ways to assess whether students can reason through a problem, respond to follow-up questions, and apply concepts under pressure. Traditional oral exams can do that, but they are difficult to scale. AI-enabled oral exams offer one possible solution.

Top takeaways

Voice AI can make oral exams more practical for larger classes

A system co-developed by Ipeirotis conducts oral exams through a voice AI agent. Students receive a personalized URL and complete the exam during a set window. The exam includes customized questions about their capstone project and a second phase focused on a case discussed in class.

Because the questioning is adaptive, the system can probe student understanding rather than simply run through a fixed checklist.
AI grading can be structured as deliberation

After the oral exam, transcripts are graded by multiple LLMs using a rubric. The models grade independently, review one another’s assessments, deliberate, and then produce a final report with evidence from the transcript. Human instructors designed the rubric and reviewed cases where the models significantly disagreed.

This structure improved grading consistency across models and gave students specific feedback on strengths, weaknesses, and areas for improvement.
Assessment design still needs human judgment

The system surfaced several design lessons:
- Examiners should ask one question at a time to avoid cognitive overload
- Voice persona matters because tone can affect student stress
- Instructors should decide how questions are assigned, rather than letting the model choose
- Adaptive exams raise fairness questions because stronger answers can trigger harder follow-ups

The bottom line

AI-enabled oral exams can help educators assess understanding in a world where written assignments are easier for a student to outsource. The strongest use case is creating assessments where students must demonstrate reasoning, respond to pressure, and show what they actually know.

What’s next?

In the age of AI-enabled exams, educators will need clear rubrics and transparent grading evidence. They’ll also need to make thoughtful design choices around stress, accessibility, and adaptive questioning.

Teaching agentic AI

The big question

How should MBA programs teach AI agent management when the technology is moving quickly?

Why it matters

AI agents are shifting from passive assistants to systems that can take actions on behalf of users. They can access tools, work across applications, run for longer periods of time, and participate in workflows that resemble organizational processes.

Future managers need to know how to design, deploy, and evaluate the products of agentic systems. And they will need to decide where human judgment belongs in the workflow.

Top takeaways

Teaching agents requires a systems mindset

Agents should be seen as business processes. A course taught by Lobel and Jagabathula focuses on how to architect an agent, build its workflow, connect it to tools, give it context, evaluate its performance, and think through deployment. The emphasis is on durable concepts that can remain useful even as specific technology changes.
MBAs need to build agents, even in a no-code environment

The instructors’ course is designed to be hands-on from the first session. The goal is to help MBA prototype agentic systems and understand how design choices affect performance.
The human-in-the-loop question is central

Humans can provide judgment, accountability, and guardrails. They can also become bottlenecks. Leaders will need to distinguish between automation workflows, where agents act independently, and augmentation workflows, where agents support human decisions.

The bottom line

Agentic AI presents a new way to think about process design and delegation in organizations. MBA students need enough hands-on experience to understand how agents are built and enough managerial judgment to know how they should be used.

What’s next?

Courses will need to move beyond tool demonstrations and focus on context, reliability, and human oversight. As agents become more capable, business schools will need to teach students how to design systems that are as accountable as they are powerful.

External CSS