The Next Generation of AI Learning Isn't About Better Chatbots — It's About Better Decisions

When a business school student pastes a case study into ChatGPT and receives a competent analysis in fifteen seconds, something real was lost. Not the output. The output looks indistinguishable from something the student could have written after two hours of work. What was lost is the thinking that the two hours would have produced. The struggle with ambiguity. The practice of weighing incomplete information. The judgment that comes from making a call and living with the consequences.

At LiveCase, we read the Ivey Business School report "From Chalkboard to Chatbot" (August 2026) as a timely and serious effort to give faculty a language for this problem. The report reviews 14 AI-enabled learning tools, but its most valuable contribution is not the list. It is the design standard beneath the list: educational AI should protect the thinking, practice, and judgment that students might otherwise outsource. That reframes the conversation from better chatbots to better learning decisions.

The report asks a better question

Not "Which model is smartest?"

Most conversations about AI in education start with the wrong question. Faculty ask which tool detects cheating best. Administrators ask which LMS integration saves the most faculty time. Vendors compete on which model scores highest on benchmarks. The Ivey report, authored by Julian Birkinshaw and Mazi Raz, starts somewhere else entirely. It asks what part of the learning process a tool is designed to protect.

That shift matters. The report evaluates tools not by their technical sophistication but by their pedagogical intent -- whether a tool preserves the productive friction that makes learning durable, or whether it smooths that friction away (PDF pp. 4-5, 9). This is not a technical standard. It is an educational one. It also complements the arguments we have made in earlier posts about decision-based assessment and moving beyond AI detection toward real engagement. The report provides the pedagogical framework those practical arguments were missing.

"What part of learning does the design protect?"

The report frames every tool through a single lens: where in the learning journey does it intervene (preparation, in-class, reinforcement, assessment) and what does it intend to develop (conceptual knowledge, skills, or judgment). A tool that holds back answers to force reasoning is judged differently from one that supplies instant explanations, even if both use the same underlying model.

This is the kind of analytical clarity that helps faculty move past the hype cycle. Instead of asking "Is this AI good or bad?" the report gives them a sharper question: "What does this design ask the student to do that they could not otherwise do, and what does it protect them from outsourcing?"

Productive friction is the central design challenge

Why struggle, practice, and consequence matter

Durable learning is not comfortable. It requires students to grapple with unfamiliar concepts, practice skills repeatedly, and exercise judgment under ambiguity. These activities create what learning scientists call productive friction: the gap between what the task demands and what the student can currently do. That gap is where growth happens.

The Ivey report makes this point directly. It warns that generative AI has made productive friction optional by allowing students to delegate effort. The immediate result may be a better output (a well-written report), but the reasoning that would have produced lasting learning is lost. The report cites research describing this as "metacognitive laziness" (PDF pp. 2-3).

How AI turns default friction into an optional choice

Before Gen AI, friction was the default. A student who had not read the case could not fake a discussion contribution. A student who had not practiced the framework could not generate a passable analysis. AI changes this equation. It makes competent output available to anyone with a prompt.

The report's insight is that this is not inherently bad. The question is whether the friction being removed was productive or merely administrative. Summarizing a long text is busywork. Deciding which information matters in an ambiguous scenario is learning. Good AI tools, the report suggests, remove the first kind of friction and preserve -- or create -- the second.

What the strongest design patterns have in common

The Ivey team identified recurring patterns across the seven public tools and seven pilot projects they reviewed. High-value tools share five design principles:

Withholding rather than supplying. The strongest tools do not answer student questions immediately. They require students to reason, to search, to ask better questions before receiving information. This mirrors the best Socratic teaching.

Counterargument rather than agreement. Tools that introduce opposing viewpoints, strategic dilemmas, or stakeholder conflicts force students to navigate trade-offs. Agreement is comfortable. Counterargument is instructive.

Information-seeking rather than instant context. Tools that present partial information and require students to determine what else they need to know build research skills and judgment. Tools that dump the full context on arrival build neither.

Rehearsal and feedback rather than one-shot performance. The most educationally valuable tools enable repeated practice with structured feedback. A single case discussion is valuable. The ability to replay the same scenario with different choices -- and see the consequences -- is transformative.

Process evidence rather than final output alone. Tools that capture how students arrived at their decisions -- their information-seeking behavior, their revisions, their time allocation -- give faculty something the final answer never can: visibility into the thinking itself.

The report notes that these principles map directly onto two pedagogical roles for AI: as a traffic cop that sequences learning activities, and as a sparring partner that challenges students' thinking (PDF p. 5). Both roles assume the student, not the AI, remains the active agent.

We have written before about the difference between chatbots and simulations and about using AI chatbots to bring case learning to life. The report's framework gives us a better vocabulary for why the distinction matters: one pattern treats AI as an answer provider, the other as a decision environment. That is the difference between a shortcut and a scaffold.

Why "better decisions" is a more useful category than "better chatbot"

From content generation to decision environments

The tools the report profiles are not all designed the same way. Some function as conversational tutors that answer student questions. Others function as decision environments where students must choose, face consequences, and revise. The report does not rank one category above the other -- it evaluates each on pedagogical fit. But its findings point clearly toward a pattern: tools that treat AI as a stakeholder, coach, or evaluator consistently demand more cognitive engagement than tools that treat AI as a provider of answers.

From AI as proxy to AI as stakeholder, coach, or evaluator

When AI plays a stakeholder role, students must question it, persuade it, and account for its interests. When AI plays a coaching role, it challenges students' reasoning rather than supplying conclusions. When AI plays an evaluator role, it assesses the quality of students' decisions, not the polish of their prose.

These are the roles the report's strongest tool designs gravitate toward. They are also roles that generic chatbots cannot easily replicate. A student cannot paste a negotiation simulation into ChatGPT and get the same experience. The scenario structure, the branching consequences, and the faculty visibility are embedded in the design, not in the prompt.

We should note here that the report explicitly identifies assessment as important but states that it is not its primary focus (PDF p. 9, note 10). Our reading extends the report's principles into the assessment domain, which is where we see the strongest alignment with scenario-based, decision-focused evaluation.

What LiveCase can add to this conversation

Scenario structure around AI interactions

The report's findings align closely with the design principles we have built into LiveCase scenarios. Each scenario places students inside an unfolding situation where they interact with AI-driven characters, gather partial information, and make decisions that produce visible consequences. The AI does not answer their questions. It plays the role of a skeptical board member, an anxious customer, or a withholding informant. This is the sparring partner role the report describes, implemented at course scale.

Decision points, consequences, and debrief

Every LiveCase scenario is structured around discrete decision points, each with branching outcomes. Students experience the consequences of their choices and see how alternative paths would have played out. The debrief surfaces not just what students decided but why and when. This creates the process evidence the report identifies as a critical design pattern.

Faculty visibility and assessment design

The report emphasizes that instructors need to see how students are thinking, not just what they produce. LiveCase provides this through analytics that track information-seeking behavior, decision timing, revision patterns, and the reasoning students submit alongside their choices. These data trails give faculty insight into the process of learning that a final essay -- particularly one that may be AI-generated -- cannot provide.

The research questions LiveCase should help institutions study

No platform should claim to have definitive answers about what works across every context. The report calls for "common measures agreed in advance so that tools can be compared substantively," and for systematic recording of what was tried, with whom, and what happened. We take that seriously. The research questions we want to help institutions explore include: Do scenario-based assessments predict real-world decision-making better than traditional exams? What conditions make AI coaching most effective for different student populations? And how much productive friction is optimal -- can we measure the point at which challenge becomes discouragement?

These are empirical questions, not marketing claims. We are running experiments alongside our partner institutions and publishing the results. We welcome any business school that wants to join that work.

Three practical questions for faculty

The report offers faculty a concrete starting point. For any course activity where AI could intervene, ask three questions:

What is being shortcut?

Identify the specific learning activity where students may outsource effort to AI. Is it the reading? The analysis? The practice? Be precise about what form the shortcut takes and what learning it bypasses.

What friction are we restoring or creating?

Decide whether your intervention restores productive friction that AI removed (for example, a closed-book component or a device-free discussion) or creates new forms of productive friction that did not exist before (an interactive simulation that forces students to determine what information they need). Both approaches are valid. The mistake is doing neither.

What evidence would show learning beyond satisfaction?

The report warns that adoption and student satisfaction are weak signals. A tool students love may be one that enables shortcuts. A better indicator is whether students are engaging with the friction or finding ways around it. The data trails from modern tools let you see this directly. Use them.

Read the complete LiveCase guide to AI-resilient assessment or join a faculty roundtable on measuring authentic learning in an AI-enabled world at livecase.com.

FAQ

What is the Ivey "From Chalkboard to Chatbot" report?

It is an August 2026 white paper by Ivey Business School professors Julian Birkinshaw and Mazi Raz that reviews 14 AI-enabled learning tools for business schools. The report evaluates tools by their pedagogical intent rather than technical sophistication and argues that AI should preserve productive friction in learning.

Which AI learning tools did the report review?

The report reviewed seven publicly available edtech tools and seven pilot projects created by Ivey faculty and staff, ranging from conversational tutors and AI coaching systems to simulations and interactive cases.

What does productive friction mean in education?

Productive friction is the gap between what a learning task demands and what a student can currently do. It is the struggle, practice, and judgment that produce durable learning. When AI removes this friction entirely, students gain efficiency but lose the opportunity to grow.

Why does the report favor focused learning tools?

The report found that tools designed to cover the entire learning journey rarely perform every part well. The most useful tools target a clearly defined learning activity with a specific pedagogical purpose.

Will AI replace classroom teaching?

The report concludes that AI tools will not displace in-person teaching. None of the tools reviewed replicated a well-run discussion where students must articulate reasoning, respond to disagreement, and learn from one another in real time. The case method may become more, rather than less, valuable.

What can simulation designers learn from the report?

The report identifies five design patterns that high-value tools share: withholding answers rather than supplying them, introducing counterarguments, requiring information-seeking, enabling repeated practice with feedback, and revealing the process of student thinking rather than just the final output.