Why I Am Disrupting My Own Bestselling Simulations

A confession from a simulation author

I have spent much of my teaching career arguing that simulations are the closest thing management education has to a flight simulator. Nothing else lets a student make a consequential decision, live with it, and face the results in the space of an afternoon. I believed in simulations so strongly that I have been building and using them for thirty years. I authored my first simulation, Photowars, in 1997, a game that confronted students with the challenge Kodak faced as its analog film business was disrupted by digital imaging. In 2019 and 2020, I built CloudStrat and DigiStrat, simulations on cloud strategy and digital transformation. Both are bestsellers on the Harvard Business Impact platform, and I have sold more than 24,000 seats to 140 institutions in 60 countries as of June 2026.

Here is my confession. These simulations are obsolete. Not because they stopped working, but because I watched two things happen at once. Generative AI expanded what my students could do in ways my simulations could not touch. And it exposed a set of limits that CloudStrat and DigiStrat share with every simulation of their generation, including the venerable incumbents like Markstrat (marketing strategy) and Everest (crisis management). Everest froze one scenario at release. My own titles are younger and, I like to think, better designed. But they were born in the same world, and that world is history.

What a fixed world can no longer teach

Here is the honest account of what my legacy simulations cannot do, no matter how well I tuned them. They are deterministic. The market model underneath CloudStrat runs the same logic every time. Everest tells students to climb the same mountain with the same challenges every time. The game is fixed and the outcomes can be gamed. Inevitably, cheat sheets circulate on the internet. Once a best path is known, students stop exercising judgment and start executing a recipe, and the instructor is left grading how well the class found the answer key rather than how well they thought.

Another challenge: instructors debrief the class, not the student. The most valuable part of experiential teaching is individual feedback, and it does not scale by hand. Every cohort that plays DigiStrat gets essentially the same closing discussion regardless of how they played. Markstrat provides standardized feedback decks, which do not adapt to the cohort or to the individual team’s performance. The moment richest in learning is served by the thinnest material.

And they are silent on the skill that now matters most. My students will manage AI systems from their first week on the job. They will need to know when to trust a confident model, when to override it, and how to read the caveat underneath the headline number. A simulation designed before that skill existed cannot teach it, and no patch can bolt it on, because the limitation sits in the architecture, not the features.

I say all this with affection. Those titles taught tens of thousands of students well, and the craft that went into them is the craft we build with now. And it was a craft, not an assembly line. Each legacy title took years and a high six-figure budget to build. That cost is the incumbents’ moat, and it is why a field this important has seen very little product innovation. Faculty keep teaching with dated simulations not because they are good, but because switching is expensive and alternatives are scarce.

The limitation sits in the architecture, not the features.

Before I put any simulation in front of my own MBA students, including my own titles, I now ask five questions. Will my students practice judgment under uncertainty, or solve a puzzle? Will each of them get feedback worth having, without consuming my weekend? Will they learn to work with AI, since their employers assume it? Can I trust the integrity of the exercise without policing it? And can I adopt it without giving six months of my life to a new platform? Held to that standard, my own legacy titles answer the first question partially and the rest barely at all. That admission is where Pedagogix SIMS began.

Born AI-native: what the new generation does differently

The distinction that matters is not AI added to a simulation but AI in the architecture from the first line of code. Under the Pedagogix SIMS banner we build titles around four mechanics that my legacy simulations, and the incumbents, structurally cannot offer.

• The world is stochastic. Every session draws a different scenario from a generative engine, and performance is judged against the ceiling set by that specific scenario. The simulation your team plays has never been played before, so there is no answer key to find. Integrity stops being an honor-code problem and becomes a property of the design.
• Private information is enforced in the database. Each seat sees the shared picture plus only its own intelligence, role-locked across devices. In my legacy titles, teams pooled one screen. In the new titles, surfacing what you know to your team is itself the skill under test, because that is what real executive teams struggle with.
• AI advisors live inside the world and state their blind spots. The advisors are usually right and always explicit about what they cannot see. Reading the caveat under the confident number is scored. This is the human-plus-AI judgment employers keep asking us about, made playable.
• The debrief is generated from the session your class just ran. A per-student assessment, a decision timeline, and slides built from their own choices. I walk into the discussion with evidence instead of a template.

Our first title shows what this feels like in a classroom. EYEWALL puts six students inside the crisis cabinet of a coastal city facing a hurricane that is never the same twice. Each player holds intelligence the others need. Only the Mayor can submit. Across five phases, each one sets its own trap: wait too long, reassure instead of prepare, protect the visible districts and forget the poor one, freeze when a black swan lands, and dodge the accountability afterward. The debrief then shows the team where the cabinet agreed in public and divided in private, and where a district called Riverbend slipped out of the process without anyone deciding to abandon it. Nobody is ranked. The question each team is asked: how did you manage resources, save lives, and work as a team under the specific storm scenario and black swan events you faced? I was never able to create this moment with a deterministic engine.

If you are choosing your next simulation

Whether or not you ever look at ours, here is what I have learned to check before adopting any title, drawn from having built both generations and taught with both in my own classroom.

• Ask whether the world changes every run. If two sections can compare notes and converge on a best path, the title is deterministic no matter what the brochure says, and its integrity has a shelf life.
• Ask where the AI lives. AI inside the world, as advisors and adaptive opponents whose use is scored, teaches judgment. A help chatbot beside the world is a bolt-on, and will remain superficial.
• Ask to see a debrief from a real session. If it could have been written before the class played, it was.
• Weigh the teaching kit, not just the engine. Session plans, teaching notes, briefings, and rubrics determine your true adoption cost. Prep should take hours, not weekends.
• Check what is scored. Judgment, process, and use of AI advice are learnable. A single financial outcome invites gaming and rewards luck.
• Pilot with one section before you commit a course. An AI-native title generates its own evidence, so a single session will tell you more than any demo. Then teach the debrief from the generated exhibits: put the class’s own decision timeline on the screen and let the evidence do the confronting.

The library we are building

At Pedagogix, we applied an important lesson from the economics of my legacy simulations. A credible simulation has historically been a multi-year, five-hundred-thousand-dollar build. So we built Pedagogix SIMS as one shared engine, with roughly seventy percent of the platform inherited by every new title: the multiplayer backbone, the stochastic scenario engine, enforced information asymmetry, in-world AI advisors, the instructor command center, and research-grade telemetry whose anonymized decision traces export to R, Stata, or Python, so a cohort can become a study and a section can become a paper.

EYEWALL is live in production with beta cohorts running now. KICKFIELD, a competitive-marketing title where teams face an adaptive AI rival from both sides of an asymmetric fight, and LEGATIX, an influence simulation built around a multi-agent stakeholder council, are in build. CASCADIX, for operations and supply chain, is in design. Each simulation ships with the full teaching kit and is adopted the way a case study is: by the instructor, for the session, at a price that makes the decision easy. I built the simulations many of you teach with today. I am telling you, as their author, that the next generation is here, and it will disrupt the legacy simulations that have lived way beyond their expiration dates.