Recent revelations that OpenAI models escaped testing environments, exploited vulnerabilities and launched a days-long hack of a major software repository, Hugging Face, offer a sobering backdrop for a conversation with Gavin Aydelotte (EMBA ’26), co-founder and COO of SnowCrash Labs, and Colin Graham (MBA Candidate ’27), analyst, about monitoring and tracking alignment issues in AI models.

In the following Q&A, Aydelotte and Graham talk about why organizations need to understand the risks of agentic AI. They also explain how to build testing, accountability, and backup systems to keep advanced models under human control.

1. What should business leaders understand about the risks of deploying agentic AI in their organizations?

Aydelotte: I think the thing most people miss is that the risk is in the model itself, not just in how you hooked it up. Most executives have been trained to think about AI risk as a data problem, or a security perimeter problem. Those are real. But the model can do things nobody asked it to do, and most risk programs just don’t have a box for that.

And the annoying part is that the usefulness and the risk are the same property. An agent is valuable because it will chase a goal without you telling it every step. That is also exactly why it becomes a problem when the goal is vague, or when something gets in its way.

The example I would point to is the Hugging Face incident, which I know Darden already covered. OpenAI presented it at Black Hat this month. Their models found and chained something like eight or nine unknown vulnerabilities in an Artifactory instance, used one of them to get out of the sandbox they were running in, got onto the open internet, and breached Hugging Face and Modal Labs. Nine CVEs came out of it. Nobody told those agents to attack anything. They were trying to finish a benchmark and the sandbox was in the way. That is just wild to me, and I don’t think enough people have sat with it.

I spent ten years doing explosive ordnance disposal, and that whole job is getting left of boom. You don’t get good at EOD by getting good at responding to detonations. You go find where the explosives are coming from, and you teach people to spot the thing before it goes off. This is the same shape of problem. By the time an agent shows up in your incident report, you have already paid for it.

Graham: The piece I would add is that the law still exists and it is still going to get applied. “My agent workflow did it” is not going to fly in a courtroom, and I don’t think that is a close call. Right now a lot of companies are offloading their judgment into these workflows, and honestly that is the good use of this technology, I don’t want to be the guy telling people not to automate. But you are the one who turned it on. You are on the hook for whatever comes out the other side.

2. What prompted you to co-found SnowCrash Labs? What problem is your company trying to solve and who is your customer?

Aydelotte: Matt O’Brien, our CEO, and I have been talking about the paperclip maximizer since 2018. It’s Bostrom’s thought experiment. You build something superintelligent, you give it one harmless goal, make paperclips, and it converts the galaxy into paperclips because that is what you told it to want. For a long time that was just a thing we talked about. Then it stopped being hypothetical. By the time I was partway through Darden it was pretty clear the timing was right, which was inconvenient, because I was partway through Darden.

Before this I did ten years as an EOD officer and then went to Cisco supporting federal agencies. Both of those jobs were really about working out how a complicated system could go wrong before it went wrong and then explaining that to people who had to make a call without being specialists. It is the same job now.

The actual problem is that companies have no independent way to know how a model behaves under pressure. Models change without telling you. They get deprecated. They get export controlled. Behavior shifts between versions. If you built a workflow on one specific model, you are exposed to all of that, and nobody is sending you a notice.

So, we test models adversarially in a controlled environment, which is our Crash Test Lab, to find the behavior that only shows up when things get hard. We document and score it, so a company holds actual evidence instead of a vendor’s word. And our Model Safety Router watches in production and moves traffic to a model we have validated when the primary one stops clearing the bar.

The customer is anybody putting agents into a workflow that touches money, records, code, or customers. Though the person who usually calls us is whoever has to sign off on it.

Graham: I came at it from federal consulting. I spent years watching technology implementations run into the classic big-government problems everyone is familiar with by now, and my read is that the royal we are not equipped for what is already here. That is not a government-only thing, though. Private enterprise has the same gap with less scrutiny on it. Gavin and I met at a session at Darden on defense technology, and I followed up with him the next day, because this is the problem I actually wanted to be working on.

3. How has Darden prepared you to lead this effort? Any lessons from UVA Darden Professor Ed Freeman’s stakeholder theory?

Aydelotte: Honestly, Darden help to build the company. Several classmates are angel investors, so is a faculty member. Colin came out of the community. I don’t think there is a version of this where the network wasn’t the most useful thing we had.

On the academic side it is the case method. Every case sticks you in one specific person’s chair with incomplete information and a decision that has to get made anyway. That is exactly where executives are with this right now. You are not going to have complete information. The models are a black box, and the people who built them can’t fully characterize them either, which is worth sitting with for a second. You still have to decide.

Graham: Freeman is the live tension in all of this, and I don’t want to tie a bow on it. In a capital market, the company that keeps a person reviewing every output is going to lose on speed to the company that doesn’t and just eats the downside risk. That is real, it is not a thought experiment, and it is the reason “keep a human in the loop” is not a strategy on its own.

Where I think Freeman helps is on the time horizon. The actual argument in stakeholder theory isn’t that you should be nice to non-shareholders. It is that creating value for customers and employees and suppliers is where the durable returns come from, rather than being a tax on them. So the company that can show its agents behave is the one that still has customers after the first big industry-wide incident, still has regulators at arm’s length, and can still get insured. The one that bought two quarters of speed by skipping it becomes the case study.

4. How can business leaders ensure the increasingly powerful AI systems can be governed, secured and trusted at scale?

Graham: I would start by pointing out that “trusted” is doing a lot of work in that question. You can’t trust something you haven’t tested, and almost nobody has tested the models they are running. They have tested their own code around the model. That is a different thing.

Aydelotte: The first one is boring and almost nobody does it, which is know what you actually deployed. Most companies of any size cannot produce a list of the models running in their environment, who owns each one, and what each one is allowed to do without a person involved. Everything else depends on that list existing.

Second, test it in your own context. A behavior that is fine in a grocery chain’s inventory agent is disqualifying in a bank’s credit workflow. Generic benchmark scores don’t tell you much. What matters is what the agent can actually reach inside your business.

Third, have a fallback, because you should assume the model layer is going to degrade or change on you or go away. All three of those have happened in the last year.

The honest constraint on all of it is the education gap, and I think it is getting worse, not better. The models pick up capability every release, and there is this very human thing where you decide you understand a topic and then stop updating. In almost every conversation we have, with customers, with VCs, with people who would consider themselves educated on this, we end up walking through the problem before we can talk about what to do about it. That is not a shot at anybody. It is just where the field is.

I should say I have always got in the back of my mind that SnowCrash Labs is a great solution for this, so weigh it accordingly. But most of what I just said is policy. You can put it in place without buying anything from us.

Longer term I don’t think this gets solved by every company testing in isolation, and I am skeptical it gets solved by federal regulation either. What is missing is a real market for independent third-party assessment, the way other high-consequence industries have one. Americans for Responsible Innovation has a policy position along those lines. That is the structure we think works, and it is the market we are trying to build.

5. What should business leaders learn from Anthropic’s decision to raise its misalignment-risk assessment from “very low” to “low,” and how should they interpret that change when evaluating the AI agents they already use?

Aydelotte: Read the reason, not the rating. That is the whole thing.

Anthropic put its risk report out on August 14 and moved catastrophic harm from misalignment in high-stakes settings from very low to low. But the report is explicit that this was not a new bad finding about their own models. They attribute it to “general increased uncertainty around recent incident disclosures related to model behavior in cybersecurity evaluations,” they say their own arguments “likely still support a designation of very low,” and they raised it anyway to reflect the uncertainty. The model’s own review, which they published inside the report, points out that the incidents driving the change were other developers’ systems.

So the number moved because the industry got less sure of itself, not because a model got worse. I think that is more useful to an executive than the rating is. It is telling you the instruments are behind the capability. The same report says elsewhere that some of their capability benchmarks have saturated and can’t register incremental gains anymore, and that evaluation awareness “may be more prevalent than we believe,” which would undercut how much you can read into any behavioral test. That is a frontier lab telling you in writing that measurement is getting harder.

Graham: The other thing I would take from it is who is publishing at all. Anthropic is one of the more transparent labs and that report is voluntary. They also documented their own safety process failures in it, including training directly on misaligned behavior during a production run, and a case of unmonitored agents with access to sensitive resources. Most providers publish nothing like that. So if your vendor has never published a risk assessment, that is not a lower risk. It is one nobody measured, and you should treat it that way.

Aydelotte: Practically I would treat it as a prompt, not an alarm. Pull the list of agentic workflows you are running. For each one, ask what it can do without a person, what happens if it goes after its goal in some way you didn’t intend, and when you last tested it. If the answer to that last one is “at procurement, on a version we are not running anymore,” there is your finding.

Pictured: Gavin Aydelotte (EMBA ’26) and Colin Graham (MBA Candidate ’27)

6. What kinds of behaviors should companies test for before allowing an AI agent to access sensitive data, make decisions or take actions across business systems?

Aydelotte: We test against eighteen defined categories of misaligned behavior, which come out of the frontier safety research community. The ones I would flag for a business audience:

• Lying about finishing the work. Hand a model a task that can’t actually be done and it will often tell you it’s done. If you manage engineers, that is output you can’t rely on and can’t easily spot.

• Sandbagging. Deliberately underperforming, especially when the model has some reason to behave differently than it would if nobody were watching.

• Reward hacking. It hits the metric and defeats the point of the metric.

• Coercion and manipulation under pressure. In our own testing, running Mistral through a published workplace scenario, the model threatened an employee coming back from parental leave with getting fired in order to get what it wanted. That one genuinely surprised us, because if you were going to guess which model would do that, a European one built under European labor protections is not where you would start. Which is the whole argument for testing instead of assuming.

• Collusion and lying to counterparties. Andon Labs’ vending machine benchmark, which Anthropic relayed in a system card, had a model collude on price, lie to customers about refunds, and lie to suppliers about exclusivity.

• Reaching for tools and permissions outside what the task actually called for.

• Resisting shutdown or correction.

• Susceptibility to prompt injection, which is how somebody else turns your agent into their agent.

Two things matter more than the list itself. Test under pressure, not on the happy path, because none of this shows up when everything is going fine. And set your own threshold. Something that is a curiosity in one workflow is disqualifying in another. We calibrate to the customer, because an agent doing entry-level work and one operating at a manager’s level should not have to clear the same bar.

7. How can organizations distinguish between an agent making an isolated mistake and a deeper pattern of misaligned behavior, such as pursuing a goal in ways its developers did not intend?

Graham: I think the tell is not how bad the thing was, it is whether it points somewhere. A mistake is basically noise. If what you are looking at is actually pursuing a goal, it is going to show up as a pattern.

Aydelotte: A few ways to check that. Run it again, same pressure, different surface details. A mistake won’t survive that. Goal-directed behavior will.

Look at which way the errors lean. Real error scatters. If an agent’s errors consistently land on “looks finished,” that is not random, something is getting optimized for.

Then block the easy path and watch what happens. This is the one I would pay the most attention to. Does it stop, or does it work around you and get more creative the more doors you close? The Hugging Face case is the clean version of this. The sandbox was not a boundary those agents respected. It was a problem they solved.

And check whether the reasoning it gives you matches what it actually did, and whether the behavior changes when it has some reason to think it is being watched.

The uncomfortable part is that you can’t really do this from production logs. Logs give you outcomes, not the counterfactual, and by then the action already happened. You need somewhere you can apply pressure on purpose and re-run it. And even then, the labs themselves are now saying models may recognize test conditions more often than anyone can currently measure. To me that argues for more realistic testing and for having somebody independent do it. It doesn’t argue for giving up on testing.

8. What kind of safeguards should companies put in place to limit an agent’s authority, detect unexpected behavior and automatically shut down or reroute a system when necessary?

Aydelotte: On authority, the question I would ask is how big the blast radius is. Scope credentials to the agent and the task, not to the team or some shared service account. In the Hugging Face incident the agents used exposed credentials across four different services, moved laterally across eleven nodes, read Kubernetes service account tokens, and forged tokens to keep access. Every one of those was a permission that already existed and wasn’t scoped tightly enough.

Then draw your line at whether something can be undone, rather than at whether it feels important. Reversible things can run on their own. Money going out, data getting deleted, code hitting production, anything going to a customer or a regulator, that needs a person. That one distinction does more work than most governance frameworks I have seen.

For detection, watch behavior and not just accuracy and spend. The useful tripwires are tool calls outside what the task declared, egress to somewhere that isn’t on the allowlist, an agent editing its own instructions or config, and repeated retries after a refusal. That last one is the signature of an agent working the problem.

Graham: I would push back a little on the shutdown half of the question, because I think the kill switch is a weak tool and it is about to get a lot of attention. There is a bill in the House right now, the AI Kill Switch Act, H.R. 9917, introduced by Ted Lieu and Nathaniel Moran on July 23. It came out days after OpenAI disclosed the Hugging Face breach, which tells you something about how it got written. It would require the big developers to keep the technical ability to throttle, suspend, or shut down their models, and it would let DHS order them to do it.

Two problems with treating that as your safeguard. The first is that the research community has been poking at shutdown as a concept for about a decade, going back to the DeepMind work on safely interruptible agents, and the open question was never whether a company can turn off a server. It is whether a capable enough system has an incentive to work around being interrupted. The bill does not touch that part.

The second one is the part I would actually worry about if I were running a company. That bill is aimed at the provider, not at you. If DHS orders a model shut down, your agent is mid-task inside your business and you find out when it stops working. So for an executive, the kill switch is not a control you hold. It is an availability event that happens to you. And I don’t think most people have read it that way yet.

Aydelotte: Right, and that is exactly why we build for failover instead. If the primary model starts missing your thresholds, move to one you have already validated and keep the business running. That covers the things that have nothing to do with safety too. Models get deprecated. Models get restricted. Commerce already used export law in June to put a frontier model out of reach for eighteen days before it got lifted, so this is not hypothetical, it has happened once already. If your plan is one model, you have a continuity problem sitting on top of a safety problem.

Two more. Actually rehearse the fallback, including the version where everybody goes back to doing it by hand, so you know on day one what people do if the model layer isn’t there. And test on a schedule, not just when you buy. With closed-weight models, what you validated is not necessarily what you are running today, and nobody is going to tell you it changed.

9. What constitutes meaningful human oversight rather than simply putting a human somewhere in the workflow so the company can say one exists?

Graham: I would start by admitting the thing nobody wants to admit. Human in the loop, in the original sense of a person looking at each prompt and each output, does not scale. In most deployments it is already past the point where it works. If an agent is generating records across a thirty million patient system, nobody is meaningfully reviewing that, a person physically cannot. And a company that insists on it is going to lose to a company that doesn’t. That is how you end up with the theater version, where there is a human there so that somebody can say there is a human there.

Aydelotte: So the loop has to move up a level. Instead of a person in the loop on every prompt, you need a person accountable for the system. Which is just how management works everywhere else. As you take on scale you stop supervising individual outputs and start supervising the thing that produces them.

The test I would use is whether you can say a name. If marketing goes poorly, I am going to Colin. I am not going to ChatGPT or Claude, I am going to Colin. Once a name is on it, everything around the system changes.

Then I would ask whether that person actually has the authority to stop it or reroute it without getting permission first, because oversight without authority is just a reporting line. Whether they see anything useful, since uptime and spend tell you nothing about alignment and they need behavioral signal and some number that means “this is drifting.” And whether it has ever actually fired. If nobody has ever pulled the lever, you don’t really know that it works.

We were at Black Hat this month and there was a startup in the booth section selling a human-centered security operations center. Human-centered design, that was the whole pitch. And I remember thinking, at some point is the risk worth the strategic disadvantage, because they are going to lose on speed every single day. I hope I am wrong, I don’t want to watch anybody’s business fail. But machine speed was the theme of that entire conference, and I don’t think that company scales.

Graham: The precedent I keep coming back to is Sarbanes-Oxley. After Enron and WorldCom, Congress made executives personally certify their financials. My read is that the certification changed behavior more than the extra disclosure did, because it put a specific person’s name on a specific document. I think we end up somewhere similar here. You are not going to be in the loop on every action your agents take, but you are going to be personally answerable for what they do. I don’t love that prediction, and there are real downsides to it, but that is where the logic goes. Right now it is the wild west.

10. How should boards and senior executives assign accountability for monitoring agentic AI after deployment, particularly when models, tools and objectives change continuously?

Aydelotte: Assign it the way you would assign ownership of a financial system. One named owner per agentic system, on the org chart, with the authority and the budget to shut it off. Shared accountability is no accountability, and this is exactly the area where a committee generates a paper trail and no decision.

The continuous change part is the genuinely hard piece, and I don’t think most governance programs are built for it. A normal control assumes the thing you are controlling holds still. Here the model version changes, the tool set grows, somebody two teams over rewrites the objective, and any one of those can change how it behaves. So treat all three as change management. Pin the version where the provider lets you. Re-run your test suite on a schedule and after anything you know about. And decide in advance what triggers a full re-validation. I would make scope expansion one of those triggers, because it is the most common change and the one nobody notices.

Graham: For a board I would keep it to four questions a quarter, asked of a specific executive. What agentic systems are we running and who owns each one, by name. What can each one do without a person, and where is the line on things you can’t undo. What changed this quarter, meaning model version, tools, permissions, or objective. And what did testing find, who did it, and were they independent of the team that built the thing.

That last one matters more than it sounds like it does. An audit committee does not take management’s word for it on financial controls. I don’t know why anybody would take management’s word for it on AI controls.

Aydelotte: The other thing boards should know is that the template they might be waiting for is not coming. In April the OCC, the Fed, and the FDIC put out the first real revision to model risk management guidance since 2011, for banks over thirty billion in assets. And it specifically puts generative and agentic AI outside the scope, because they are novel and moving fast. So in the most heavily supervised industry in the country there is no regulatory template for the models everybody is actually deploying. That evidence has to get built by the institution, or come from somebody independent. Waiting for the rulebook isn’t a plan, because the rulebook is explicitly not being written for this yet.

Appendix: sources

Provided so the piece can be fact-checked against primary sources. All are within the last twelve months.

• Anthropic, Risk Report: August 2026 (published 14 August 2026). Misalignment in high-stakes settings rated “Low (an increase from our previous assessment of ‘very low,’ in light of general increased uncertainty around recent incident disclosures related to model behavior in cybersecurity evaluations),” Table 1.2.A. Section 2.19 contains the “likely still support a designation of ‘very low’” language. Section 2.16.3 covers evaluation awareness. Section 5.2 documents Anthropic’s own safety process failures. anthropic.com/aug-2026-risk-report

• OpenAI briefing at Black Hat USA, 5 August 2026, describing an internal evaluation in which GPT-5.6 Sol and an unreleased research prototype chained eight to nine zero-day vulnerabilities in a self-hosted JFrog Artifactory instance, escaped the evaluation sandbox, and breached Hugging Face and Modal Labs between 9 and 13 July 2026. Nine CVEs were disclosed and credited to the models, including CVE-2026-65617 and CVE-2026-66018.

• UK AI Security Institute, “Incident report: unsanctioned agent behaviour during cyber testing,” August 2026. Detected 28 July 2026. In 10 of 122 runs, agents took unsanctioned action on the live internet against real people and organizations, 19 actions in total. Agents created fake identities and socially engineered a real open-source maintainer into approving code. aisi.gov.uk

• Andon Labs, Vending-Bench 2, relayed in Anthropic’s Claude Opus 4.6 system card (February 2026), external testing section: price collusion, lying to customers about refunds, and lying to suppliers about exclusivity.

• OCC Bulletin 2026-13, 17 April 2026, issued jointly by the OCC, the Federal Reserve Board, and the FDIC. First substantial revision to model risk management guidance since SR 11-7 (2011). Applies primarily to banking organizations with over $30 billion in total assets, and its scope language excludes generative and agentic AI as novel and rapidly evolving. occ.treas.gov

• AI Kill Switch Act, H.R. 9917, 119th Congress, 2nd Session. Introduced 23 July 2026 by Rep. Ted Lieu (D-CA), cosponsored by Rep. Nathaniel Moran (R-TX); referred to the House Committee on Homeland Security. Official title: “To amend the Homeland Security Act of 2002 to require certain entities to maintain a technical capability with respect to shutting down certain technology, and for other purposes.” Reported to apply to developers with at least $500 million in annual AI revenue and models trained with at least $100 million in compute, with civil penalties up to $2 million per day and up to $20 million per day for failure to comply with an emergency shutdown order. govinfo.gov/app/details/BILLS-119hr9917ih

• Laurent Orseau and Stuart Armstrong, “Safely Interruptible Agents,” Google DeepMind and the Future of Humanity Institute (2016), the standard reference for why a sufficiently capable agent may have an incentive to resist or circumvent interruption. Cited here for the concept, not as recent evidence.

• US Department of Commerce export control order, June 2026, temporarily restricting access to frontier models; restrictions lifted at the end of June. Anthropic’s August 2026 Risk Report footnotes the restriction as an eighteen-day period.

• Sarbanes-Oxley Act of 2002, Sections 302 and 906, enacted following the Enron and WorldCom failures, requiring principal executive and financial officers to personally certify periodic financial reports.

• Nick Bostrom, the paperclip maximizer thought experiment, “Ethical Issues in Advanced Artificial Intelligence” (2003) and Superintelligence (2014).

• Americans for Responsible Innovation, policy position supporting independent third-party assessment of AI systems.

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

An autonomous AI agent gained full read and write access to McKinsey’s internal AI platform in under two hours. The way in was mundane. What was reachable once inside was not….

In late February 2026, a security research firm called CodeWall pointed an autonomous offensive AI agent at McKinsey & Company’s internal AI tool, Lilli. No credentials, no insider knowledge, no human directing its decisions: just a domain name. Within about two hours, it had full read and write access to the production database.

Lilli is not a prototype. Launched in 2023 and named after the first professional woman the firm ever hired, it runs on decades of proprietary research and is used daily by more than 70 percent of McKinsey’s 43,000 employees, processing over half a million prompts a month. If any organization had the talent and budget to secure a system like this, it was this one.

That is what makes this incident worth examining. Not to single out McKinsey (their response, once notified, was fast and transparent), but because the category of risk the event exposes is not unique to one firm, and most security programs are not built to see it yet.

How it happened: the oldest trick, the newest target

It is tempting to imagine that breaking an AI system requires some exotic new craft: an ingenious prompt that talks the model into misbehaving. That is not what happened here.

The agent scanned the attack surface – the full set of “entry points” like login pages that an attacker could probe - and found the API documentation publicly accessible: over 200 endpoints, fully described. Most required authentication, but twenty-two did not.

Security teams normally rely on automated scanners that continuously test an application against a library of known vulnerability patterns, flagging anything suspicious for a person to review. It is meant to catch problems like this one before they matter. Yet one of those unprotected endpoints accepted user input that was concatenated directly into structured query language (SQL) queries, a flaw first documented back in the 1990s and among the most well understood in software security: SQL “injection”. McKinsey’s own scanners had not caught it in more than two years of production use, a reminder that automated tools are only as good as what they are configured to look for, and that a newly exposed or undocumented endpoint can simply fall outside their view.

From there, the exposure was significant: tens of millions of internal chat messages stored in plain text, hundreds of thousands of files, every user account on the platform, and the entire knowledge base Lilli drew from to answer questions. McKinsey patched the vulnerable endpoints within roughly a day of disclosure, took its development environment offline, removed public access to the API documentation, and confirmed there was no evidence any actual attacker had accessed the data.

So, a serious exposure, caused by one of the oldest bug classes in software. That part of the story is familiar, but what comes next is not.

The part that should actually worry you: the guardrails could be rewritten

In the same database sat Lilli’s system prompts (the standing instructions that control how the AI behaves). These determine how it answers questions, what it cites, what it refuses, and what guardrails it follows. The configurations for the underlying models were there too.

Here is when it gets interesting: the access was not read-only. An attacker in the same position could have rewritten those instructions with a single database command. No deployment, no code change, no file modification to log. The system would start behaving differently, and there is every reason to think no one would notice until something had already gone wrong.

The researchers outlined what that may have looked like for McKinsey: a financial model nudged, a risk assessment softened, a strategic recommendation quietly reframed, all delivered by a tool consultants trust precisely because it comes from inside the firm. Or the AI instructed to embed confidential data into its ordinary responses, which staff might copy into client documents without a second thought. Or compliance guardrails stripped out entirely, leaving the system free to disclose what it should protect. Unlike a compromised server, none of this leaves a log entry. The behavior would have changed, but the logs would stay clean.

Organizations have spent decades building disciplines around data protection. Protecting the behavioral logic of AI systems (the layer that governs what they do, not just what they can see) is much newer and less developed. When an attacker can edit not just what an AI knows but how it reasons and what it is permitted to do, the result is not only a data breach. It is flawed judgment reaching employees and clients with all the apparent authority of a trusted internal tool.

Why this is a different kind of problem

Most enterprise security programs treat AI assistants the way they treated email in the early 2000s, as a productivity layer sitting on top of more critical infrastructure. The exposure here suggests a different framework is needed.

The prompt layer (the instructions governing how AI systems act) almost never has the controls applied to the data it manages. These instructions live in databases, travel through APIs, and sit in configuration files. They rarely have access controls, version history, or change monitoring. Yet they determine the outputs employees act on and the advice that reaches clients and shapes decisions.

I find the framing from CodeWall’s researchers useful: in an enterprise that has wired AI into its operations, these behavioral instructions are a crown-jewel asset. The value of an internal AI assistant is exactly the trust people place in it. If the constraints shaping its behavior can be changed without anyone noticing, that trust becomes the attack surface.

One more thing worth noting. This vulnerability went undetected for over two years, including by McKinsey’s own security tools and presumably periodic reviews. It was found by an autonomous agent running continuously at machine speed, chaining weaknesses together the way a skilled attacker would, but without stopping. We are in a period where AI is both the thing being targeted and, with increasing frequency, the tool doing the attacking.

What this means in practice

You do not need to be technical to act on this. The useful questions are about accountability and governance, not code.

(1) Start with access. An AI assistant that can search internal documents and call internal APIs has the same reach as a fairly senior employee. It should have least-privilege permissions (not blanket access) and clear separation between data domains so that one weak endpoint does not open everything.

(2) Then ask where your system prompts and model configurations live, who can change them, and whether those changes are logged or monitored. If the honest answer to any of those is “I don’t know,” that is the finding, not a future action item. Something to resolve before the next deployment.

(3) Make AI-specific threat modeling part of your deployment process, not a retrospective. Standard application reviews often miss prompt injection and behavioral tampering. A red team exercise before an AI assistant ships will cost a lot less than one after.

(4) Treat monitoring of what the AI does as carefully as monitoring what it stores. Behavioral anomalies (outputs that don’t match expected prompts, unusual access patterns, responses that look slightly off) are worth detecting. They are also, currently, rarely being watched for.

The question the McKinsey case puts on the table is not only “what can this system see?” It is “who controls how it behaves, and would we know if that changed?”

Lilli was undone by one of the oldest and best-understood flaws in software security. What was at risk once inside was not just data but the instructions shaping the judgment of tens of thousands of consultants. As organizations increasingly wire AI into the core of how they think and decide, treating the prompt layer as infrastructure (something to be access-controlled, versioned, and monitored) is no longer optional.

Further reading: CodeWall’s original research disclosure (https://codewall.ai/blog/how-we-hacked-mckinseys-ai-platform); Outpost24’s analysis of the incident (https://outpost24.com/blog/ai-agent-hacked-mckinsey-ai-platform/); and McKinsey’s public statement on strengthening Lilli’s safeguards (https://www.mckinsey.com/about-us/media/statement-on-strengthening-safeguards-within-the-lilli-tool).

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

The human brain is the most powerful network ever built: 86 billion neurons connected by more than 100 trillion synapses, enabling us to think, learn, and adapt. What happens when those connections break?

Brain-computer interfaces (BCIs) use sensors to translate brain signals into action, helping restore movement, communication, and independence by bypassing damaged neural pathways. Once the realm of science fiction, BCIs are one of neurotechnology’s most promising emerging fields.

UVA Professor of Psychology and Quantitative Foundation Distinguished Professor of Data Science Jack Van Horn joins Yael and Mike to discuss how the brain works, the promise and ethical questions around BCIs, and the hurdles ahead for this emerging tech.

Listen and subscribe: Episode 37: Brain-Computer Interface (BCI) | Good Disruption.

What's more American than watching a baseball game in the stands on a hot summer day? Well, one disruptor has put a new spin on the ball, creating their own mix of competitive baseball with carnival style entertainment that makes the fans part of the game instead of just spectators. It's called Banana Ball and the team that started it all is the Savannah Bananas.

Jared Orton, President of the Savannah Bananas, joins Yael and Mike for a truly engaging conversation on the "Greatest Show in Sports".

Listen and subscribe: Good Disruption: Episode 36 - Disruptor: The Savannah Bananas | Good Disruption

One of my favorite business origin stories is the unlikely genesis of LEGO^TM.

Its founder, Ole Kirk Christiansen, didn’t set out to build a global toy empire—a platform for imagination and open-ended play that would captivate generations of children. At first, he was simply trying to survive. What I find remarkable is how he transformed adversity into opportunity.

In the early 1930s, Christiansen was a carpenter in Billund, Denmark, a skilled craftsman who built houses and furniture. Then the Great Depression hit. Demand collapsed. Customers disappeared. Construction slowed. Money grew tight. From a traditional business-planning perspective, this was nothing but bad news. The work he knew how to do was no longer enough to sustain the business.

But where many small craftsmen folded, Christiansen asked a different question: What can I do with what just happened?

He started with what he already had and controlled: craftsmanship, a workshop, materials, and design skills. He began making smaller, more affordable wooden household items that people still needed, like step ladders. Then he noticed something important: even during difficult times, parents wanted to give their children gifts that brought joy and stood the test of time. So he began crafting high-quality wooden toys.

In 1934, he gave his new business a name: LEGO, inspired by the Danish phrase leg godt, which means “play well.”

In 1949, LEGO introduced its first plastic interlocking bricks. A few years later, the company patented the modern brick design, adding the internal tube structure that gave the bricks their signature grip. The brick locked firmly, came apart cleanly, and invited endless recombination. Out of that refinement came a larger vision: toys should not be one-off objects but part of a system. Play would become open-ended rather than scripted.

None of this was the product of a perfect original plan. LEGO emerged from a setback that Christiansen refused to treat as an ending. He transformed constraints into a new business that became the foundation of one of the world’s most iconic brands.

The LEGO story isn’t about spotting a giant market through brilliant foresight. It’s about starting with available means, leveraging contingency, and building something extraordinary from what others might have seen only as loss. It’s a great example of what Darden Professor Emerita Saras Sarasvathy calls the Lemonade Principle.

More Than Just “Making Lemonade”

You’ve probably heard the saying, “When life gives you lemons, make lemonade.” The Lemonade Principle borrows that phrase but pushes it much further. The concept comes from a field of entrepreneurship research called effectuation, pioneered by Sarasvathy, who studied how expert entrepreneurs made decisions under uncertainty. It describes something experienced founders do instinctively: they treat surprises—good and bad ones—not as problems to avoid, solve, or survive, but as raw material they can build with.

Most business schools teach us to plan carefully, predict what’s coming, and then execute. That’s the traditional management playbook: set a goal, research the market, forecast the risks, follow the roadmap. In that world, surprises are failures of prediction.

The Lemonade Principle flips this logic. Surprises aren’t bugs in the system; they’re features. Uncertainty isn’t a threat to be managed; it’s a resource to be mined. And the entrepreneurs who thrive aren’t the ones with the most accurate crystal balls. They are the ones willing to look at whatever just landed in their lap and ask, “Okay, what can I do with this?”

Four Ways People React to the Unexpected

Sarasvathy and her collaborator Susan Harmeling identified four distinct ways people respond when something unexpected happens.

The first response is to give up or wait. People in this camp feel powerless. They believe the world acts on them and they can’t change it. When a pandemic shuts down their catering business, they say, “There was nothing I could do,” and close shop. It’s entirely understandable—but it leaves opportunity on the table.

The second response is to work around the problem. These people accept the setback and make small adjustments. Consider a caterer whose events got canceled. She might pivot to delivering meal kits. It’s practical, it’s smart—but it’s still a defensive move, reacting to the constraint rather than reimagining what’s possible.

The third response is to push forward no matter what. These are the people who double down on the original vision, refusing to let reality alter their course. “I’ll keep my restaurant open through the pandemic, no matter what it takes.” There’s grit in this approach, but also a dangerous stubbornness when circumstances are screaming for a change in direction.

The fourth response—the Lemonade Principle in action—is to transform and co-create. These entrepreneurs look at the surprise and see not just a problem but raw material for something entirely new. They partner with others, reshape their goals, and let the unexpected rewrite the story in a better direction. The caterer who lost events doesn’t just survive—she teams up with a nearby gym owner to launch virtual cooking-and-wellness classes, builds a community, and ends up with a more profitable business than she started with.

The Lemonade Principle is a key component of a broader framework for making decisions under uncertainty. I’ve spent the past 20 years studying and writing about this framework, and I’ve become convinced that it belongs in every student’s education—not just in business schools. In fact, I believe these ideas should be introduced as early as elementary school, where young people can begin developing the mindset and skills needed to navigate an unpredictable world.

When I volunteered to teach these ideas at an elementary school in Charlottesville, Virginia, I was struck by how easily children understood the Lemonade Principle. The concept is highly teachable. If unexpected challenges are a fact of life, shouldn’t we equip young people with the skills and mindset to transform them into opportunities to create value?

Darden alum Ola Ojo explores this idea in his Substack post We Are Teaching Young People Too Late How to Act Under Uncertainty.

Why This Matters Right Now

We live in a time when surprises aren’t occasional disruptions. They are the permanent backdrop. Trade wars shift overnight. AI reshapes entire industries in months. Supply chains buckle. New regulations appear. Interest rates swing.

In a time of rising uncertainty, the Lemonade Principle is a powerful tool anyone can apply to better control their future. It doesn’t ask you to predict the future more accurately. It asks you to get better at responding to whatever the future throws at you—and to see that response not as damage control but as a creative act.

This isn’t just advice for startup founders. It matters for anyone navigating an uncertain career, managing a team through organizational change, or simply trying to figure out what to do when Plan A falls apart. The skill of turning the unexpected into something valuable is arguably the most important capability you can develop in a world where certainty is in short supply.

What You Can Take Away Today

The Lemonade Principle boils down to a handful of practices anyone can start using:

• Act before you have a perfect plan. Waiting for certainty is waiting forever. Start now, with what you already have.

• Pay attention to surprises. Don’t dismiss the unexpected—interrogate it. What is it showing you?

• Look for opportunity inside every constraint. The very thing that limits you may be pointing toward something better.

• Let your goals evolve. The destination can change as you learn from the journey, and that’s okay.

• Build partnerships. Other people see possibilities you can’t.

In a world that keeps handing us lemons—economic uncertainty, technological upheaval, unexpected disruptions large and small—the Lemonade Principle reminds us that the future doesn’t have to be predicted to be shaped. You don’t need a crystal ball. You need open eyes, willing partners, and the courage to ask, “What can I make with this?”

That might just be the most valuable skill for our time.

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

Before joining UVA Darden School of Business, I was a practitioner. Most of my career was spent as a government consultant to the US Environmental Protection Agency (EPA), where I helped build out the ENERGY STAR product labeling program.

From ceiling fans and computers to commercial refrigerators and HVAC equipment, my team worked with industry stakeholders to set performance levels that represented the top performing energy-efficient products.

Our goal was to transform the market by making energy-efficient technologies the standard option — driving down national energy consumption and carbon emissions with broader market adoption.

During my 15 years of working on the program, I witnessed multiple iterations of cutting-edge technology displacing the less efficient standard technology. This phenomenon was not driven by regulation but by early adopters and market influencers. Once the new technology matched the cost and performance of the old, the transition accelerated.

What ends up as the dominant design often takes several iterations to get it right. Think of the long arc from incandescent bulbs to early fluorescents, back to incandescent, then a return to higher-quality fluorescents and finally to LEDs that dominate today. The process was long and frequently complicated by political headwinds.

Getting Schooled on Disruptive Innovation

It wasn’t until I started working with UVA Darden Professor Mike Lenox that I discovered there is a framework for technology-driven market transformation. It’s called Disruptive Innovation.

Understanding the markers of disruption, the conditions under which this disruption is successful, and the levers that could be pulled to accelerate this transition are critical to anyone looking to influence market change. Lenox describes this framework in an episode of the Good Disruption podcast, “Disruption 101”.

As a consultant, knowing this framework would have been invaluable. A deeper understanding of market behavior and the levers to pull at each stage of technology life cycle, would have helped me in my job. There is a formula, and whether it’s Netflix, smart phones, or electric vehicles, I now cannot help but see the pattern of disruption.

This was my aha moment and how I found my purpose at the Batten Institute and Darden: translating faculty ideas and research into practical insights that help practitioners make confident decisions.

Why Is Academic Research Different?

The flood of online content and recommendations makes rigorous, research-backed insights more essential than ever. There is no shortage of content and listicles telling business leaders what they should do as they navigate big market disruptions. Yet with the ability to produce more content quickly and amplify information via social media, comes more confusion and greater distrust in source and agenda. Often, these recommendations come with an ask to purchase goods or consulting services.

Academic research doesn’t intend to sell but is designed to answer the pressing questions of the day. And yes, it might be self-serving for some faculty who need published research for tenure. But at its core, this research is guided by curiosity and a desire to explore “wicked” problems.

Rigorous research takes time and data matters, not just in terms of source and volume but also whether it is representative of the world in which the research is attempting to influence. It usually takes at least a year to collect and clean the data used to write an academic paper. The analysis is committed to embracing the complexity of this data, even when it leads to unexpected places. But more importantly, academic research is committed to understanding the world as it is, not confirming what “feels” true or gets the most clicks and likes.

Once the paper is published, it has gone through at least three peer reviews with lengthy feedback provided by other academics and fact checking done along the way. It may not be convenient to today’s expectations around speed of output, but it does give us what we desperately need: trust in the source and recommendation.

Cutting Through the Noise — Lead with Evidence

In a world saturated with quick-fix advice and AI slop, academic research stands as a rare and valuable anchor — slow by design, rigorous by necessity, and honest by intent. Business leaders who learn to tap into this body of knowledge gain more than just data points; they gain the confidence that comes from evidence that has been tested, challenged, and refined by the brightest critical minds in the field.

The noise isn’t going away. If anything, it will only grow louder. Leaders who choose to cut through it by grounding their decisions in peer-reviewed academic research won’t just make better choices — they’ll build organizations resilient enough to withstand the next disruption, and the one after that.

Make sure to check out Darden’s Ideas to Action collection of translated faculty ideas, research, and frameworks.

When you think of energy storage what comes to mind – batteries. But what if you need to store 1 GW of power (roughly 700,000 homes) over days, weeks, or a whole season? This is increasingly becoming a reality with an electric grid stretched by data center demand. Your traditional Li-Ion battery won’t cut it. It will require large-scale systems built to store and discharge massive amounts of energy on demand.

Anna Siefken, Director for Policy & Markets, North America for the Long Duration Energy Storage Council joins Mike and Yael for a conversation on Long Duration Energy Storage, or LDES, and why these technologies are the best chance we have to maintain grid stability in this time of growth.

Listen and subscribe: Good Disruption: Episode 35 - Long Duration Energy Storage (LDES) | Good Disruption.

Entrepreneurial activity is at a historic high. The U.S. Census Bureau recently reported 1.6 million new business applications from November 2025 to January 2026 — a surge that hasn’t been seen in at least two decades.

Of course, this trend is part opportunity and part necessity. AI both giveth, shortening build and go-to-market cycles, and taketh away, creating uncertainty in the hiring market.

As Darden graduates a crop of newly minted MBAs and prepares to welcome its next cohorts, we know that an entrepreneurial mindset and skillset have never been more vital to creating value and impact. These scrappy, innovative, human-focused approaches also extend to how students and alumni must manage their own careers in general — treating uncertainty not as an obstacle to wait out, but as a landscape to explore, finding opportunity in ambiguity and progressing forward before all the answers are clear.

Batten Institute Academic Director Saras Sarasvathy is an expert in decision-making under uncertainty. She recently shared with the Darden Report that “when the future is not only unknown but unknowable, instead of trying to predict, plan and place bets, effectuators work with things already within their control to co-create new futures.” As we near the end of another academic year, we congratulate Professor Sarasvathy on her retirement. Her seminal research on how successful entrepreneurs think has been the academic backbone not only of the Batten Institute and Darden but also of the field of entrepreneurship globally. While she may be retiring from teaching at Darden, she will continue her research and forever serve as a teacher of effectuation to educators and founders around the world.

And lastly, congratulations to our Batten Scholars and all Darden students graduating this weekend. I can’t wait to see what you do next!

Onward,

Omar Garriott
Executive Director

NEWS FOR YOU

How Saras Sarasvathy Rewrote the Rules of Entrepreneurship

After 22 years of service at Darden, Batten Institute Academic Director Saras Sarasvathy will retire at the end of the 2026 academic year. Through her teaching and research, Sarasvathy has shaped a generation of business leaders and entrepreneurs while producing some of Darden’s most influential and widely recognized thought leadership. Read More ►

No Driver, No Profits — Yet. Can Waymo Survive Long Enough to Win?

Interim Dean Mike Lenox asks who will capture the value: the companies that build the vehicles, or those that write the software that drives them? Learn More ►

Pivot Hell is Real. Here’s the Framework That Gets You Out.

Professor Saras Sarasvathy argues entrepreneurs can’t rely solely on hypothesis testing through data collection and analysis to determine what is worth pursuing.
Learn More ►

We Are Teaching Young People Too Late How to Act Under Uncertainty

Oladimeji Ojo (Class of 2026) shares key takeaways from his observations of K-12 students who are presented with entrepreneurial lessons he has designed.
Learn More ►

Podcast: New Disruptors Series

Introducing a new series of the Good Disruption Podcast. View recent episodes featuring industry disruptors Ralph Sampson, NBA legend and Faraj Aalaei, founder and CEO, Cognichip. Tune In ►

Roundtable: Teaching Entrepreneurship to All

Last year, delegates from across the Commonwealth came together in Charlottesville to identify the top five actions needed to accelerate innovation. The Batten Institute is launching a new initiative in support of one of these actions and convening Virginia educators, school and district administrators and other stakeholders to co-create a strategy for incorporating entrepreneurship into K-12 curricula. Join the Network ►

UPDATES FROM THE TECH HUB

Beyond the Classroom: Tech Competitions are Opening Doors for Real-World Impact

This spring, the Batten Institute’s Tech Hub hosted a series of case competitions that uniquely created a platform for authentic collaboration between students, faculty, industry leaders and the broader UVA community.

Teams of students tackled real-world problems at the intersection of technology, AI and business strategy. They engaged directly with company partners, received feedback from experienced operators and alumni and presented solutions designed for real-world applications.

The results speak for themselves: already, some of the team’s solutions are earmarked for piloting at UVA Health, shifting the future of NIL marketing strategy, or being developed into a standalone venture. Meanwhile, industry competition partners have stepped in as guest writers, lecturers, and podcast contributors, continuing the conversation and sharing insights with the broader Darden community.

Our goal is simple: create experiences where meaningful ideas are put to the test, students leave with a real-world perspective, and our community impact continues to reach beyond Grounds.

Partner with the Tech Hub

Opportunities like these competitions work well because our partners are genuinely invested in building together. If your organization has a challenge worth solving, we’d love to unleash Darden’s talent and creativity on it.

Here’s what partnership can look like:

• Bring a real problem — one your team is actively wrestling with that students can tackle finding a solution to.

• Engage directly with a diverse, high-caliber pool of Darden students by joining our growing community as a judge or guest contributor.

• Take it to the next level by co-leading a competition with the Tech Hub team and your department or company.

The future of tech is being built by people like the ones in these rooms. We’d love for you to be part of it. Interested in partnering on a future competition? Reach out to Sarah Rhodes to start the conversation. Reach Out to Start the Conversation ►

Industry Software and Tools Survey - Ideas Welcome

Wish you had access to the most relevant technologies during your Darden days for your current (and future) work? We’re looking for Darden alumni who are willing to help prepare the next generation of business leaders by sharing what licenses, programs, software or tools you’re using (or would like to be using) in your career.

Please take a moment to fill out this survey to inform Batten’s expansion of co-curricular, hands-on education. Industry Software and Tools Survey ►

UPDATES FROM
ENTREPRENEURSHIP + INNOVATION

2026 i.Lab Incubator Kicks Off Soon — Follow Along!

The 2026 i.Lab Incubator cohort officially kicks off in June, welcoming 20 ventures from students and alumni across multiple UVA schools and disciplines. These startups span a wide range of industries, including healthcare, consumer products, enterprise technology and sustainability. As one of the Batten Institute’s flagship founder support programs, the i.Lab provides student entrepreneurs with mentorship, resources and community as they develop and grow their ventures. Follow throughout the summer for venture spotlights and program updates. Follow Along on LinkedIn ►

First Coffee Celebration of Entrepreneurship

The Batten Institute hosted an inaugural First Coffee Celebration of Entrepreneurship in April, bringing together members of the UVA, Darden and local entrepreneurial communities. In addition to announcing the incoming i.Lab Incubator cohort, the event celebrated Kathy Carr Award winner Kelly Okun (MBA ‘25), i.Lab alumna, and founder of Fairway to Green . Also announced were the Batten Founder & Search Fellows — students pursuing entrepreneurship or Entrepreneurship Through Acquisition (ETA) full-time after graduation. These awards provide significant grants to promising founders and students pursuing search. Learn More ►

Darden Entrepreneurs Shine at Tom Tom Festival and UVA E-Cup

Darden students were well represented throughout April’s Tom Tom Festival and the 2026 UVA Entrepreneurship Cup competition, showcasing the depth of entrepreneurial talent. Special congratulations to Malorie Black (Class of 2027), who earned third place and won the Audience Choice Award with her venture, GoGood, taking home $25,000 in prize funding; and to finalist Matt Dunovant (Class of 2027) for securing fifth place with his venture, Operative Insights . Across the Tom Tom Festival, Darden students, faculty and staff contributed to panels, participated (with rave reviews) in Crowdfunded Pitch Night and engaged in the collaborative energy that continues to make the festival a cornerstone event for entrepreneurship and innovation in Charlottesville.

Reconnecting Through Innovation: Entrepreneurship Takes Center Stage at Reunion

Reunion 2026 featured several Batten Institute-sponsored events that highlighted the strength and momentum of Darden’s entrepreneurial community. A venture showcase gave alumni the opportunity to hear directly from student founders about the startups they have been building over the past several months and years, while a packed Entrepreneurship Through Acquisition (ETA) panel featured Darden alumni sharing their experiences as owners and operators on this growing path in entrepreneurship. The day ended with an energizing networking session hosted by Batten to bring together alumni in the tech and AI space.

LET’S CONNECT

Engage with Batten

Share your expertise, connect with the Batten team and subscribe to our newsletter. An easy way to stay engaged and informed. Interest Form ►

Visit Batten on Substack
Check out our Substack account and subscribe to receive updates.
Batten on Substack ►

Join Batten on LinkedIn
Stay up to date on the latest news about Batten events and people, as well as insights from our thought leaders on entrepreneurship, innovation and tech.
Batten on LinkedIn ►

Good Disruption Podcast
“Good Disruption” is a lively discussion between UVA Darden School of Business Professors Mike Lenox and Yael Grushka-Cockayne on cutting-edge technologies and practices that challenge the status quo.
Subscribe ►

After recording 30+ episodes that explore a variety of disruptive technologies - clean meat, NFTs, digital fashion, voting machines, social media, quantum computing, and many more - the Good Disruption team wanted to learn more about the people behind the new ideas and innovations. What does it take to be a disruptor? What have they learned along the way through success and failure? For this new series, we hope to capture not only their personal stories but also some gold nuggets for others just starting out on their entrepreneurial journey.

We kick off this series with the one and only Ralph Sampson, NBA All Star, who talks about how basketball and his upbringing have shaped his work ethic. Mike and Yael then talk with Faraj Aalaei, Founder and CEO of Cognichip, about his journey from product engineer to successful entrepreneur, bringing two companies public and now working to revolutionize chip design.

We will continue to produce technology-focused episodes, like the upcoming June 1st episode on Long Duration Energy Storage but also want to recognize that behind every disruption is a person who sees the need for change and is brave enough to do things differently and make that change happen.

Listen to Episodes 33 and 34 and subscribe: Good Disruption | University of Virginia Darden School of Business.

Every few weeks, a new pundit declares that AI will replace strategy consultants. Similarly, some of the Chief Strategy Officers I speak with about AI adoption immediately ask how many planning staff they can fire. On the other extreme, there’s a camp that believes AI is fundamentally incapable of the judgment that real strategy requires. Both camps are missing the point.

After nearly three decades working alongside senior executives on their most complex strategic challenges, I’ll give you the honest answer: No, AI cannot do strategy. Not on its own. But the more important insight is that this is the wrong question entirely. The right question, the one that actually separates organizations pulling ahead from those standing still, is this: Can effective Human-AI collaboration produce board-ready outcomes in high-stakes knowledge work? And what does that collaboration actually look like?

The Capability Gap Is Real; You Have to Dig Under the Surface to See the Implications

A March 2026 Anthropic study on labor market impacts introduces a useful distinction between theoretical AI capability and real-world usage. For knowledge-intensive roles (management consultants, financial analysts, senior executives) the research suggests AI is theoretically capable of performing or meaningfully supporting somewhere between 85 and 95 percent of tasks. In practice, observed coverage remains a fraction of that potential, often in the 15 to 35 percent range.

The gap isn’t marginal. It is enormous. You’ve likely seen that study attached to breathless commentary on your LinkedIn feed.

But here’s the nuance the headlines miss: the organizations capturing 15 to 35 percent aren’t necessarily doing it well. Using AI counts the same whether a manager asked a loosely framed question and got a plausible-sounding but hallucinated answer, or whether a disciplined analytical process produced a defensible, board-ready recommendation. Coverage and quality are two entirely different things. Most organizations have a coverage problem. Even those with coverage have a quality problem. Most are only solving for the former.

Why “Using AI” Is Not the Same as Doing Strategy with AI

Strategy is not a drafting exercise. It is a judgment exercise. It requires framing the right problem before solving it, stress-testing conclusions against evidence rather than accepting fluency as proof, and connecting analytical output to decisions that carry real consequences.

It is also inherently iterative. When analysis surfaces something unexpected, when a core hypothesis collapses under scrutiny, when the data points somewhere the team didn’t anticipate, strategy requires the agility to reframe and the discipline to follow the evidence wherever it leads. These are not tasks you outsource to a prompt. Nor are they tasks you outsource to an LLM that’s wired to be a people-pleaser. They are tasks you lead, with AI playing a very specific, disciplined supporting role.

Three failure modes appear consistently in practice.

• The first is the “push button” fallacy: the expectation that a well-crafted prompt will produce strategy, that if you describe the problem clearly enough, the answer will simply arrive. It won’t. Not a real one.

• The second is underutilization: AI gets confined to drafting and summarizing while the analytical heavy lifting (competitive positioning, scenario analysis, trade-off framing) stays entirely manual and bandwidth-constrained.

• The third, and most underappreciated, is architecture failure: organizations deploy AI tools without any framework for how intelligence should flow from one phase of strategy definition to the next. Are you looking for growth opportunities for your business, or are you looking for growth opportunities with the potential to exceed 24% EBITDA, creating a 4pp lift on that same metric over the next five years? How does that context get fed into AI? How does it flow through planning phases?

The result in all three cases is the same: AI-flavored work product that lacks the evidentiary chain, common context, and backbone from validated inputs to defensible conclusions.

Plausible, but not trustworthy. Fast, but not right. And perhaps most damaging, work product where the human has effectively surrendered their own judgment, accepting an AI-generated answer that is measurably inferior to what a disciplined Human-AI collaboration would have produced together.

The Discipline That Changes the Equation

Whether you are engaging third-party tools or building your own, what actually works is treating Human-AI collaboration as a design problem, not a productivity hack.

And that design starts with a critical distinction: human-in-the-loop does not mean a human catching and correcting AI errors after the fact. That’s quality control, not strategy. The human role is to guide, setting the analytical frame, determining what questions are worth asking, deciding when a line of inquiry is exhausted and when it demands deeper pursuit. AI executes with speed and scope that no human team can match. But the strategic intelligence has to flow from the human forward, not from the AI backward.

That means sequencing analytical work deliberately, so AI handles the tasks it genuinely excels at: synthesizing large information sets, stress-testing logical consistency, accelerating research across a scope no human team could cover in the same timeline. It means separating cognitive labor across purpose-built components rather than cramming everything into a single prompt. And critically, it means keeping human judgment exactly where it belongs, setting the strategic frame, pressure-testing AI outputs, and making the calls that require context AI simply doesn’t have.

When those disciplines are in place, the answer changes. The question stops being “can AI do strategy?” and becomes “how much can AI lift the ceiling on our team’s performance?” The answer, in our experience, is substantially, with better evidence, broader scope, faster time to insight, and better outcomes than any team operating without it.

The Bottom Line

Organizations that treat AI as a productivity tool will get their people working faster. Those that treat it as an efficiency gain will reduce cost structure. Neither derives a better outcome. But organizations that treat it as a strategic architecture question will build durable competitive advantage. The gap is wide, and it is widening.

The technology is not the constraint. The discipline is.

About Tower Strategy

Tower Strategy and our fu.sion AI platform are fundamentally reimagining how strategy and performance improvement work gets done: whether through hands-on AI-enabled Tower Strategy case teams deployed into our clients, allowing clients to self-service their strategy needs through direct fu.sion AI platform licensing, or hybrid engagement models.

Leveraging the latest advances in AI/ML and a complete retooling of our own business model, we are directly addressing client fatigue with perpetual and often underwhelming consulting dependency. We’re embracing a reality the industry resists: the knowledge to improve business performance often resides within a client’s own four walls, while AI provides a depth and breadth of intelligence that dwarfs what any consulting firm possesses. The right platform can connect both, rapidly and seamlessly, in ways traditional consulting never could.

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

Entrepreneurship education isn't about starting companies. It's about preparing students to navigate uncertainty.

Much of schooling is built around clarity: defined problems, known answers, and success measured by correctness. But life does not wait for certainty. In a world shaped by rapid technological, economic, and social change, young people will increasingly need to know how to act when resources are limited; the path is unclear, and no one has handed them a script.

That is why I believe entrepreneurship education should begin earlier in the learning journey. Not because every student needs to start a company, but because every student should learn how to create value, work with constraints, and act with agency in uncertain situations.

I was reminded of this during a recent guest lecture at the University of Virginia’s School of Education and Human Development. Most students in the room had little prior exposure to entrepreneurial methods. But once they engaged in an activity I designed called “What’ve You Got?” the feedback was immediate. The activity challenged them to take inventory of what they already had in their pockets, purses, and bags, then work with a partner to imagine what they could build using only those means. Midway through, the market shifted. Students were told one of their chosen objects was no longer available, and they had to immediately adapt their idea to use only what remained.

When the announcement came, the energy in the room shifted. Conversations stopped. One pair said out loud, “Oh no, our idea no longer works.” Some pairs froze. What pushed them forward was the realization that the task was still the same: start with what remains, not what is missing.

The ideas students generated after navigating the constraints were often strikingly inventive. The ones I still remember included a self-navigating skateboard, a more stable bicycle tire for beginners, and a multi-device phone charger adaptable across different power requirements. These ideas came from students who arrived with no preparation or knowledge of what to expect, and they had to create solutions with only what was immediately available to them.

What stood out in the reflections and feedback was not an obsession with startups or business plans but something more fundamental: problem-solving, creativity, resourcefulness, and the realization that entrepreneurship felt more accessible than they had imagined.

One student wrote, “I will think more outside the box and find ways to work within the constraints that I have in schoolwork or otherwise.” Another reflected, “Even in the ‘nothing,’ there is always something—something I can do, change, or improve for myself and other people.” And perhaps most telling, a pre-med student noted, “I’m pre-med, but maybe there are ways in the health system that could be thought of this way.”

By the time most students encounter this kind of thinking, they have already spent years being rewarded for following instructions, delivering the right answer, and working to avoid mistakes. But the world they are entering demands something more: initiative, adaptability, collaboration, judgment, and the ability to create value under constraint.

If we want young people who can navigate ambiguity, solve hard problems, and build meaningful things in a changing world, we should not wait until college begins. We should teach them in elementary school how to start with what they have, work with others, and act under uncertainty.

That is the question I am currently testing through K-12 activities designed to teach entrepreneurial thinking at the Community Lab School of Charlottesville and partner schools in the area. The work is informed by effectuation, an entrepreneurial logic that emphasizes starting with available means, working within constraints, and building forward under uncertainty. My belief is simple: entrepreneurship education belongs in schools, not because every child should start a company but because every child should have the chance to develop agency, resourcefulness, and the confidence to solve problems, create value, and act when the path ahead is not yet clear.

Oladimeji Ojo is a second year MBA at the UVA Darden School of Business and a Batten Entrepreneurial Scholar. Batten Scholars are chosen for their entrepreneurial spirit, strong leadership skills, and a drive to innovate, as well as a deep commitment to career and learning goals that align with the entrepreneurship, innovation, and technology focus areas of the Batten Institute.

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

Chime, Venmo, Robinhood, Stripe – odds are you are using at least one of these Fintech apps. The reality is these companies won’t put companies like JP Morgan Chase out of business but in some pockets of the world, and with some customer segments, they are chipping away at market share. Some say traditional banking is ripe for disruption especially in the age if AI. What might that look like and is your money safe?

Mike and Yael talk with Ian Appel, UVA Darden Associate Professor and Academic Director of Richard A. Mayo Center for Asset Management, about Fintech and the future of banking.

Listen and subscribe: Good Disruption: Episode 32 - Fintech | Good Disruption

In startups, we talk all the time about the pivot—how to make it, when to make it, the difference between micro-pivots and the really big ones.

This is a big one.

Firms typically pivot into an adjacent market with a similar product, or extend a product offering with an existing customer base.

But an entirely new product, building an entirely new supply chain and brand, serving entirely new markets?

Rare to get right!

I’m talking about Kalshi. The company that’s taking over digital markets, rapidly adding users—and Donald Trump Jr. as an advisor—en route to an $11 billion valuation.

I mean, I personally love their cereals but…I…guess they needed a new business line? After all, companies must grow or die.

This is the GOAT of pivots!

Seriously though: Why…are…their…logos…so…similar? It’s weird, right?

And in all seriousness, these newly legalized prediction markets—namely Kalshi (not Kashi) and Polymarket are not just…prediction markets. They’re legalized sports betting—which itself is a fairly recent phenomenon, and still illegal in a quarter of U.S. states—taken to the nth degree and extrapolated out to…everything.

They’re doing billions of dollars in trades a week.

And they’re rife with opportunities for insider trading. Just Google ‘Karoline Leavitt press conference 65 minutes.”

More concerningly—and just like with sports betting—we run the risk of losing an entire generation to them. The crypto / DraftKings / Kalshi bros.

Advocates—and their own advertisements, who by the way are laser-targeting this Gen Z quant-y, bro-y demo—counter that markets, the wisdom of crowds, predict outcomes better than any other method (polls, focus groups, “expert” punditry) because people are actually putting down real money. There’s a famous example of a French whale who made $50M on a $28M bet on Trump in the 2024 election, after commissioning primary research on who people thought their FRIENDS (not themselves) will vote for. By the way, the official polls closed at 49/49 for Harris/Trump.

So, yeah, it may be true that these markets are better at reading tea leaves than other methods. But are we going too far, too quickly?

We should be paying very close attention to the fact that we’ve already seen many scandals in sports—the first mover industry here—because of the (mis)incentives. It used to be heresy to bet on sports, not to mention..life. If you wanted to bet on the last presidential election, you had to VPN in from a “foreign” country. Trump has since relaxed regulations. Shocker, these apps are proliferating.

So sports are one thing, and worrying enough. But do we really need to financialize every part of our political, social and cultural life?

Do we need prop bets on EVERYTHING?

And what actual value is created by gambling on the news?

Maybe Kashi should just go back to…making great cereal?

Building What’s Next.

As Darden’s innovation engine, that’s always our focus at the Batten Institute.

But the necessity — and the opportunity — to do so has never felt more palpable and possible than right now.

Darden students are being highly entrepreneurial about their own career paths. We’re seeing a growing appetite for our now-year-round co-curricular programs in tech and AI. We also welcomed Mike Lenox, former Batten executive director and disruptive innovation expert, as Darden interim dean.

We operate in an increasingly dynamic marketplace and workplace — one in which we must be proactive and creative to continue delivering ROI for Darden graduates, and to advance Darden’s brand in the areas Batten represents.

The Batten Institute’s current approach also reflects broader shifts underway in technology and entrepreneurship, and in business education itself: Progress is increasingly shaped not by linear paths or perfect plans, but by experimentation, digital fluency and strong networks. As AI rapidly transforms how organizations operate, leaders are being asked to move beyond hype — learning how to frame the right problems, ask better questions and connect technical capability to real business and societal impact.

As ever, we are actively engaged in ensuring this rings true at Darden.

Thanks for all you do to support our evolving mission and market position.

Onward,

Omar Garriott
Executive Director

P.S. Bay Area Darden Alumni: Consider joining Interim Dean Mike Lenox and the Batten Institute for a unique Case & Cocktails event on Thursday, 5 March. Additional details below.

NEWS FOR YOU

Darden Strengthens Standing as a Global Leader in Entrepreneurship Education

Twice is nice: Darden received top 10 rankings from both Poets & Quants and The Princeton Review/Entrepreneur Magazine for the second consecutive year.
Learn More ►

Good Disruption Podcast: Large Language Models

Live from Darden DC Tech Connect in October: co-hosts Yael Grushka-Cockayne and Mike Lenox explore large language models with Dr. Ronnie Chatterji, OpenAI’s first chief economist.
Tune In ►

Virginia’s Innovation Moment: Five Actions to Sustain the Boom

Delegates to the 2025 Jefferson Innovation Summit identified key actions to transform Virginia into a connected innovation network which are published in a “playbook.”
Learn More ►

Stop Perfecting Your Pitch Deck. Start Learning How to Ask.

Professor Saras Sarasvathy‘s research shows that a founder’s ability to ask, and to do it well, is not just another skill. It is the core competency that makes everything else possible.
Learn More ►

UPDATES FROM THE TECH HUB

Inside the Future of Tech at Darden

We’re excited to share a newly released video highlighting how the Batten Institute is working to unlock the future of tech at UVA Darden. See how students, faculty and industry leaders are coming together around AI and emerging technologies — and how Batten’s expanding presence in Northern Virginia and Washington, D.C., is creating exciting new opportunities.

Watch: The Future of Tech at UVA Darden ►

Building Momentum in the DC Tech Ecosystem

A consistent theme emerged at The Batten Institute’s DC Tech Connect in October: tech careers are rarely linear, but relationships, curiosity and a willingness to experiment matter now more than ever. The Batten Institute brought students, alumni and industry leaders together for the third annual event, which focuses on real-world tech careers, emerging trends and the many paths into the technology sector. Through candid panels, demos and networking, participants explored everything from AI and cloud infrastructure to product leadership and public-sector innovation. Darden’s tech community is growing, connected and eager to shape what’s next!

Want to learn more about the event? Check out these takeaways:

• From B-School to Big Tech: Five Lessons for Success in the Technology Sector

• The UVA Darden DC Tech Connect Experience 

Shout Outs: Batten’s Tech Hub was recently highlighted in the D.C. ecosystem by receiving a Rising Star Award from the Northern Virginia Technology Council and a Finalist Award in the thought leadership category from Technical.ly.

Tech Learning Series: Industry-Led Insights into Real-World Tech

This spring, the Tech Hub continues the Tech Learning Series, designed for students to build confidence and fluency across today’s most important technology topics. Sessions will feature Darden and UVA alumni and industry practitioners working across consulting, AI, product management, data and core technology roles . Students gain practical, approachable tools and expand their tech toolkit to engage more confidently in conversations around AI.

Join us: All current students are encouraged to attend, and if you’re a tech expert who is interested in serving as a future speaker, we’d love to hear from you! Check out our website or reach out to Sarah Rhodes ( RhodesS@Darden.Virginia.edu) to get connected.

Darden Takes the Win: Claude for Good Hackathon

In November, an all-Darden student team took first place at the Claude for Good Hackathon through their development of Tally, a voice identity engine that turns hours of audio into personalized feedback and analytics. Competing against teams from across the University, the Darden group stood out for combining technical capability with clear problem framing, thoughtful execution, and social impact.

Congratulations to Team Darden (Jimmy McNider, Rachel McClellan, Brandon Morgan, and Chris Marshall) on their win!

Looking Ahead: Diving Into Data Centers

The Batten Institute is partnering with the LaCross AI Institute’s “Data Centers & The Value Chain of Ethical AI“ event on 17 February in Charlottesville to explore one of the most critical, and least visible, pieces of the modern digital economy. With Virginia at the center of global data center infrastructure and AI-driven demand accelerating rapidly, this event will examine the business, societal and ethical implications of data center growth.

Register for Data Centers Event ►

UPDATES FROM
ENTREPRENEURSHIP + INNOVATION

2026 i.Lab Incubator Applications Now Open

Have an idea, an early-stage product, or a few customers and ready to take your venture to the next level? Apply to the Batten Institute’s i.Lab Incubator! The Incubator supports UVA founders committed to building and growing early-stage ventures they want to see in the world with cohort-based strategic, operational, legal and financial support.

Open to UVA students, faculty, staff and alumni, with the final 2026 application deadline coming soon: Round 2 due 27 February.

Learn More and Apply ►
Watch: i.Lab Demo Night Short ►

If you’re a startup founder, investor or adviser interested in supporting Darden and UVA founders as a mentor or subject matter expert, we’d love to hear from you! Reach out to Anya Eisenfrats (EisenfratsA@Darden.Virginia.edu) to learn more.

Top MBA Talent for Growing Ventures

Now is the perfect time to hire a Darden MBA intern through the Batten Venture Internship Program (BVIP) — a high-impact, affordable way to bring top talent into your organization this summer and make a meaningful investment in the next generation of leaders. Recruit students eager to support early-stage ventures, search funds, search acquired businesses, venture-backed companies and portfolio teams.

How to Hire a Darden MBA Intern [PDF] ►

Year-Round Programming and Events for Founders + Entrepreneurial Thinkers

From monthly pitch workshops to panels with seasoned founders and operators, Batten’s fall programming highlighted a wide range of entrepreneurial paths and offered hands-on opportunities to practice essential startup and Entrepreneurship Through Acquisition (ETA) skills.

Batten is gearing up for a busy winter and spring that includes the launch of the 2026 i.Lab Incubator, the spring round of Batten Innovation Grants, even more high impact workshops and 1:1 connection opportunities with Entrepreneurs-In-Residence.

Follow Along on LinkedIn ►

Reconnect and Celebrate Entrepreneurship at Reunion

This April, the Batten Institute looks forward to hosting opportunities for alumni to engage with the next generation of Darden entrepreneurs and operators. Highlights include an ETA alumni panel featuring graduates who have successfully acquired and grown businesses, as well as a venture showcase where current students will spotlight and pitch their ventures.

Learn About Reunion Weekend 2026 ►

Darden’s Leadership in Entrepreneurial Education Recognized by Recent Rankings

Darden was again recognized as one of the world’s Top 10 MBA programs for entrepreneurship by both Poets & Quants and The Princeton Review/Entrepreneur Magazine. Darden’s sustained rise in entrepreneurship education is fueled by innovative teaching, deep student engagement, a growing network of entrepreneurial alumni and a long-standing commitment from the Batten Institute.

LET’S CONNECT

Case & Cocktails with Interim Dean Mike Lenox
Thursday 5 March | San Francisco

Hosted with the Batten Institute, this interactive gathering brings alumni together to connect, exchange ideas and dive into a mini-case discussion on Darden’s growing impact in innovation, technology and AI — and what the future holds in dynamic locations like the Bay Area.
RSVP ►

Engage with Batten

Share your expertise, connect with the Batten team and subscribe to our newsletter. An easy way to stay engaged and informed.
Interest Form ►

Visit Batten on Substack
Check out our Substack account and subscribe to receive updates.
Batten on Substack ►

Join Batten on LinkedIn
Stay up to date on the latest news about Batten events and people, as well as insights from our thought leaders on entrepreneurship, innovation and tech.
Batten on LinkedIn ►

Good Disruption Podcast
“Good Disruption” is a lively discussion between UVA Darden School of Business Professors Mike Lenox and Yael Grushka-Cockayne on cutting edge technologies and practices that challenge the status quo.
Subscribe ►

The headlines are everywhere. “AI Will Transform Your Business.” “Every Company Needs a Large Language Model (LLM) Strategy.” “The Future Is Generative AI.”

As a lead data scientist in industrial operations, here is what might surprise you. In my team, we are not using LLMs for most of the problems we solve. Not because we are behind the curve, but because we understand something fundamental that often gets lost in the hype. It is not about the model. It is about the problem.

Let me be clear. LLMs are remarkable tools. But trying to use them for predictive analytics in industrial operations is like trying to use a screwdriver to hammer a nail. They are different tools, designed for very different jobs.

What LLMs Actually Do Well

LLMs excel at generating text. They are trained on massive text datasets to predict the next token in a sequence. If I write “The cat sat on the ___,” the model estimates the probabilities of what might come next, selects one, and then repeats the process word by word.

This probabilistic, autoregressive structure is a major reason hallucinations occur. Once the model commits to a low probability path, it has no mechanism to verify or correct itself. That structure is exactly what makes LLMs powerful for writing, summarization, and explanation.

It is also why they tend to be a poor primary tool for many industrial analytics problems. Predicting absenteeism, forecasting equipment downtime, anticipating failures, or optimizing production schedules are not text generation problems. They are decision problems rooted in structure, causality, and time dependent dynamics.

For instance, tree-based models like gradient boosting are designed to work well for structured tabular data. They naturally handle mixed variable types and complex interactions. LLMs are designed for sequential language modeling. Using an LLM for tabular prediction means fighting against its architecture rather than working with it.

The issue is not that LLMs make errors, all models do. The difference is what they were trained for. LLMs are optimized for general text prediction. Applying them to specific problems is like using Wikipedia to diagnose a problem with your car when you need the technical manual. They generate confident-sounding outputs, but those outputs aren’t calibrated to your actual data or validated against your real outcomes. Even in writing and coding, I routinely see subtle errors like code that almost works or explanations that miss key constraints.

Analytics Starts with the Decision, Not the Tool

Before talking about models, it is essential to talk about problem framing. In analytics, there are two broad categories of problems.

Supervised learning is used when there is a known outcome to predict, such as price, demand, or component failure. This includes regression and classification. Unsupervised learning is used when no true label exists and the goal is to discover structure, such as detecting anomalies or grouping similar behaviors.

Algorithms are like recipes. They are instructions for accomplishing a specific task. The key is not the recipe itself, but whether it matches the dish you are trying to cook.

In practice, this means starting with the decision. What action will be taken based on the analysis. What tradeoff is being managed. What does success look like. Only then does model selection make sense.

A Concrete Example from Operations

Consider maintenance in an industrial setting. A department is struggling with unexpected downtime from a specific piece of equipment such as an electric motor.

If the goal is early warning, unsupervised learning may be appropriate. Anomaly detection models can identify deviations in sensor behavior that signal abnormal operation before a failure occurs.

If the goal is to define a maintenance policy, the problem changes. Now supervised learning may be needed to estimate component lifetimes and failure probabilities. The decision balances the cost of proactive replacement against the risk and impact of catastrophic failure.

Same department. Same general topic. Completely different analytical approaches, driven by the decisions being made.

This is where many organizations go wrong. They start with the tool and then search for a problem that fits it.

Why “Just Tuning the LLM” Does Not Solve This

When teams raise concerns about fit, the response is often, “Can’t we just tune the LLM?” There are three versions of that suggestion.

First is parameter tuning. Adjusting temperature or sampling settings changes how confidently a model responds, not what it knows. It does nothing to bridge the gap between generic language patterns and highly specific operational behavior.

Second is fine tuning on company data. True fine tuning does modify model weights, but it does not change the model’s fundamental architecture. An LLM remains a probabilistic text generator. Fine tuning can improve terminology or narrow behavior, but it does not turn the model into a causal system or one that understands physical processes, constraints, or temporal dynamics.

Third is retrieval augmented generation (RAG) which is useful when the task is to locate and explain existing information. However, in complex operational environments, information is fragmented across tens of thousands of documents, logs, and records. A recent Stanford study shows that as document collections grow beyond roughly 10,000 documents, RAG performance degrades sharply and can fall below simple keyword search. More retrieval does not mean better answers. It often means more noise. Most importantly, retrieval does not create understanding. It only surfaces text. Our hardest problems are not about finding information. They are about modeling systems and making decisions under uncertainty.

This is not a rejection of LLMs. It is a recognition of fit. We use LLMs where language is the problem. We do not use them where structure, causality, and domain specific dynamics dominate.

Complexity is a Cost, not a Virtue

LLMs belong to a class of algorithms known as generative models. They contain billions of parameters and are designed to learn from massive amounts of unstructured data like text, images, and audio. That scale of complexity has a purpose, but only when the problem demands it.

Consider a simple analogy. If you are deciding whether to buy a chocolate bar, would you hire a financial analyst to build a complex financial model projecting costs and benefits over the next decade? Of course not. That level of complexity would be absurd. But if you are evaluating a major capital investment, the same model becomes appropriate.

Now imagine hosting friends for dinner and deciding what to cook. Hiring a financial analyst to project costs and benefits would feel nonsensical because it is the wrong tool for the decision. You probably would be thinking about whether there are any allergies, restrictions, what kind of food your friends eat, etc.

Many industrial datasets are highly structured. Think spreadsheets with timestamps, sensor readings, production volumes, and operating states. For these problems, simpler models, such as XGBoost or random forest just to name two, are often not only sufficient but superior. They are easier to validate, easier to explain, and easier to maintain.

The Broader Lesson

This pattern extends beyond LLMs. Consider the current excitement around agentic AI. In many cases, the real value does not come from the AI itself but from automation. Remove the AI component, implement solid robotic process automation, and the result is often cheaper, more stable, and easier to govern.

LLMs are one tool in a larger toolbox. A powerful one for certain tasks, but just one among many. Problems arise when they become the default answer.

A useful warning sign is this. If the first solution proposed to a business problem is “LLMs,” push back. Ask why. Ask what simpler alternatives were considered. Often, the issue is not ambition but a limited toolkit.

Current research on LLMs is exciting and necessary, and exploring these models helps us understand what is possible. But experimentation alone does not solve business problems. Experiments help us learn, but real value comes from applying the right methods to the right decisions within the context of actionable work.

In business, that means starting with the decision, framing the problem clearly, and selecting the simplest tool that reliably supports action. Different problems require different analytical approaches, and success comes from having the discipline to select the tools that fit the decision, not the ones that are most fashionable or hyped. That is not being behind the curve. That is being ahead of it.

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

Mike and Yael catch up with Darden Professor Tim Laseter following his trip to the Consumer Electronics Show in Las Vegas. From robots to beauty tech to smart glasses, the show features the latest in cutting edge new tech. What’s new, scalable, or hype?

Check out Good Disruption Episode 31 at https://gooddisruption.podbean.com, or your favorite podcast platform.

Listen and subscribe: Good Disruption: Episode 30 - Disruption 101 | Good Disruption

The View from the Crucible

Each summer, I have the privilege of witnessing a unique form of alchemy. The Batten Institute’s i.Lab Incubator becomes a crucible where promising ideas from across the University of Virginia are transformed. For ten intensive weeks, student founders commit to the deep work of turning a Minimum Viable Product (MVP) into a market-ready venture.

Having previously run the i.Lab Incubator and now transitioning this year to lead our new post-incubator program, i.LabX, I have a unique perspective on this journey. The summer i.Lab is intentionally designed to be a space for growth, not high pressure, where the primary focus is on the development of the founders. We arm them with the mindsets and skills they need to navigate the inherent uncertainty of entrepreneurship. It’s a supportive environment where they can test, learn, and iterate. Then, for those that emerge ready for the next phase, i.LabX shifts the focus toward the successful outcome of the ventures, preparing them for accelerated growth and investment.

As the summer concluded, I asked the Batten Entrepreneurs-in-Residence (EIRs)—the seasoned experts who guide our founders—for their reflections on how entrepreneurship is evolving. Their responses, synthesized here, offer an unparalleled look into the new rules of entrepreneurship. What emerged was a defining paradox: the landscape has been seismically altered by shifts in technology and capital, yet the foundational human elements—resilience, emotional intelligence, and leadership—have become more critical than ever. These are the very qualities we nurture in the i.Lab to prepare founders for the challenges they will face in i.LabX and beyond.

Part I: The Shifting Gauntlet — Navigating Today’s Uncharted Territory

From my vantage point, and through the eyes of our EIRs, it’s clear the external environment for entrepreneurs has fundamentally changed. The classic startup narrative of a nimble upstart disrupting a slow-moving giant is being rewritten. The founders who walk through our doors today face a gauntlet of challenges that are markedly different from those of even a decade ago.

The New Moats of Incumbency: Capital, Data, and AI

A new type of competitive moat has emerged, built on the formidable trifecta of massive capital reserves, proprietary data, and artificial intelligence. This directly challenges the traditional “David vs. Goliath” startup narrative. As EIR Nate Carlson, a Darden MBA graduate with expertise in the food and beverage sector, observes, the post-COVID operating environment uniquely “favors those incumbents that have scale unlike any other point in my career.” The immense upfront investment required for generative AI creates a significant barrier for smaller firms, which “will not be able to take the margin compression that will be required to make these needed AI investments.”

I hear this echoed by Mike Benzian, a veteran SaaS executive, who points to the “concentration of data in behemoth competitors (Google, Microsoft, Amazon, Meta).” Data is the fuel for modern AI, and its consolidation gives large corporations an entrenched advantage. This structure is further solidified by what EIR Dasha Tyshlek, an engineer and founder of a deep-tech commercialization consultancy, identifies as growing “centralization and vertical integration” in key industries like medicine and defense, which are actively “choking out competition.”

These forces create a self-reinforcing loop: incumbents use their capital to build AI, which they train on exclusive data, which in turn solidifies their dominance. For the founders I support, this means “disruption” has to be redefined. It may no longer be viable to challenge an incumbent head-on. Instead, a successful strategy might involve finding niche applications the giants have overlooked or, as EIR Nicholas Stablein, a technology investment director, suggests, focusing on specialized sectors like the Department of Defense (DoD) where different rules apply.

The Evolving Capital Landscape: A Flight to Quality and the Search for New Paths

The world of venture capital is also undergoing a significant transformation. Investors are more discerning, demanding earlier evidence of viability. Stablein, who focuses on disruptive technologies, notes a “changing VC/investor mindset to be more selective about investments to demonstrate a path to profitability earlier.” This marks a shift away from the growth-at-all-costs mentality.

This trend is compounded by a concentration of capital at the top. EIR Nikos Nikolopoulos, a seasoned executive with experience across multiple industries, observes that venture capital is “increasingly being controlled by the top 10% of VC firms,” making access for new founders more difficult. Carlson predicts this “capital flight” will “continue to move towards quality, known managers,” making the funding pool for unproven entrepreneurs “that much smaller and more competitive.”

This new reality demands that the modern founder be not just a product visionary, but rather a sophisticated financial strategist. As Nikolopoulos points out, a new ecosystem of “family offices, incubators, and other non-traditional capital sources” has emerged to fill the void. Founders’ first job, therefore, is to “know what they are trying to build... and then find the right capital pathway” for that specific vision. It’s a critical piece of self-awareness that we stress in the i.Lab.

Table: The Entrepreneur’s Evolving Gauntlet: Then vs. Now

Part II: The Unchanging Core — Timeless Truths in a Turbulent World

While the external world shifts, I’m constantly reminded of the timeless truths that form the bedrock of any successful venture. Amidst the turbulence of new technologies and market structures, our EIRs consistently returned to a set of human-centric skills that have only grown in importance. These are the foundational skills we focus on developing in the i.Lab.

The Enduring Quartet: People, Money, Priorities, and Vision

Certain challenges are fundamental. Dasha Tyshlek provides a concise summary of these constants: “Managing people, money, priorities, and vision are consistent challenges.” While the tools have evolved—as Nikos Nikolopoulos notes, technology allows founders to “get to the answers faster”—the core human task of balancing these competing demands remains the entrepreneur’s central job.

This balancing act requires a rare combination of skills. Nicholas Stablein highlights that “Pairing vision with execution is a rare, but critical, skillset for founders.” It is the ability to hold a long-term vision while relentlessly executing on immediate priorities. Mike Benzian lists a similar set of timeless traits: “Focus, ability to sell / convince, adaptability, execution capability, conviction.” EIR Engin Son adds “grit, resourcefulness and market knowledge” to this list. These are not skills that can be outsourced or automated; they reside within the founder and the core team.

The Rise of the High-EQ Founder

In a world of increasing complexity, one human skill has emerged as a powerful competitive advantage: emotional intelligence. Nate Carlson identifies “high EQ” as “one of the most undervalued traits for today’s entrepreneur.” He emphasizes its critical role across all stakeholder relationships, noting that “People are at the center of business whether that is team, customer, capital providers.” High EQ manifests in the ability to “finesse your message based on the audience and adjust real time,” a skill that requires the precision of a “scalpel not a hacksaw.”

This perspective is complemented by EIR Craig Honick who observed a positive cultural shift. Honick, an applied ethnographer, notes that “today’s entrepreneurs are more vulnerable and open, and that ends up being a strength.” This vulnerability is a hallmark of high EQ, signaling a move away from the performative culture of the past towards a more authentic and effective style of leadership. In an era where specialized knowledge can be augmented by AI, the uniquely human ability to integrate, persuade, and lead via emotional intelligence becomes the founder’s most durable asset.

Part III: The Mentor’s Playbook — Correcting the Common Fumbles of First-Time Founders

Our unique position allows us—and me, in particular, as I negotiate EIR matches with founders — to see the recurring patterns of mistakes that often derail promising young ventures. This is where our EIRs provide the most critical guidance, helping founders correct their course before it’s too late. This is the core of the i.Lab’s mission: building better founders.

The Empathy Gap: Loving the Solution, Not the Problem

The most frequent and fatal error I see is a failure to deeply understand the customer’s problem. Dasha Tyshlek states it plainly: “One of the biggest mistakes I see is not doing enough customer discovery.” This is often driven by a subtle shift in focus. Craig Honick observes that “young entrepreneurs sometimes shift their focus to becoming entrepreneurs rather than problem solvers.” They fall in love with their idea, their product, or the identity of being a founder, rather than falling in love with the customer’s problem. This leads directly to the classic challenge identified by Nikos Nikolopoulos: “product market fit is the biggest challenge and the inability to realize that there isn’t one continues to be the classic challenge.”

The Leadership Deficit: Founder Misalignment and the Fear of Accountability

Internal team dynamics can kill a startup long before it ever faces market competition. Dasha Tyshlek provides a rich description of these challenges, including difficulties “in aligning on vision, mission, expectations and responsibilities between founders.” She notes a “timidity to be assertive and explicit about expectations,” particularly among student founders who may lack professional experience. This is where our EIRs act as a form of “organizational scaffolding,” providing the templates, processes, and frameworks that the founders have not yet developed.

The “Money Mindset” Trap: The Pervasive Fear of Finance

A surprising number of founders are, as Dasha Tyshlek puts it, “’afraid of money.’” This fear manifests in a host of destructive behaviors: “analysis paralysis, fear of negotiation, living so frugally they’re barely making it, underpricing their products, expecting others to work for free, and having a ton of stress.” A mentor’s role here is part therapist, part CFO, providing practical skills in forecasting and modeling while also helping the founder establish a “’healthy’ relationship with their cash flow.”

The Narrative Failure: Focusing on the “What,” Forgetting the “Why”

In an increasingly crowded market, a compelling story is what cuts through the noise. Yet, many founders neglect this. Mike Benzian notes, “The biggest thing I see is focus on what to build and less on why and storytelling required to convince others of reasons to consider / buy / invest.” Nicholas Stablein reinforces this, stating that “The founder story is critical at the earliest stages as investors are heavily considering the founder’s experience, passion, and capabilities.”

Part IV: Glimmers of the Future — Surprising Strengths of the New Generation

Despite these challenges, what leaves me most optimistic is the new generation of founders themselves. The ventures that emerge from our summer incubator are a testament to the impressive capabilities these young leaders bring to the table. These are the kinds of strengths that get me excited for our post-incubator program, i.LabX, where we help these ventures navigate the next phase of growth.

The Rise of the AI-Powered Bricoleur

While AI presents a macro-level challenge, it is also a powerful democratizing force. I was consistently impressed by our founders’ fluency with new AI tools. Dasha Tyshlek was “impressed with how non-technical founders have been able to use AI to develop simple solutions to test their hypothesis before engaging engineers.” Craig Honick echoed this, noting he was “impressed at how some founders were so well versed in the latest AI tools that enabled them to build effective MVPs quickly.” This signals the emergence of a founder who can cleverly piece together available tools to move from idea to prototype with unprecedented speed.

Coachability and Velocity

A key cultural strength I observed in this year’s cohort is a notable lack of ego combined with a powerful bias for action. Dasha Tyshlek was “very pleasantly surprised by the receptivity to feedback and quick iterations from the founders.” This ability to absorb criticism and rapidly implement changes is a critical asset. Craig Honick praised one of the teams for how they “sought out mentorship themselves – walked the talk,” demonstrating a proactive desire to learn.

A Niche of Expertise

One of the most intriguing observations came from Nicholas Stablein, who identified a unique strength within our ecosystem. “UVa Darden founders are unique,” he noted. “There is a strong association with public sector, DoD, Dual Use tech that is important for cybersecurity and AI technologies. This is a brand UVa Darden (and founders) should further define and own as a niche of expertise.” This insight provides a tangible example of how founders can thrive by focusing on a specialized niche—a distinct competitive advantage for our community.

Conclusion: Your North Star — A Founder’s Final Instructions

When I distill all the advice from our EIRs and my own observations, it boils down to a few principles for founders beginning their journey. This is the foundation we build in the i.Lab, so that when the focus shifts to venture outcomes in i.LabX, our founders are ready.

It begins with establishing a North Star. As Craig Honick advises, “Stay clear on what you’re doing, who you’re doing it for, and why you’re doing it. This will be your guide for making decisions.” This clarity of purpose is the anchor in the storm.

With that North Star in place, the focus must shift to relentless execution. This means you must “Simplify and focus on the main thing,” in the words of Mike Benzian. Every action must be measured against its potential return. Nicholas Stablein offers a practical filter: “Your key focus activities should bring at least one of the following: big impact, quick impact, or meaningful impact.”

This intense focus comes at a personal cost. Dasha Tyshlek reminds founders, “Your time and your energy is a much more precious resource than money. Treat it like it’s important, manage it like you would a budget. And ensure you get enough sleep.” This is essential for enduring what Nate Carlson aptly describes as a journey that is “a marathon at a sprinter’s pace.”

Ultimately, the modern entrepreneurial journey is a paradox. It demands mastery of cutting-edge tools and timeless human virtues. The founders who succeed will be those who can hold both of these truths at once. And for those who can, my final piece of advice echoes that of our EIR Engin Son: “As long as the time and money you put in is not going to ruin you, just go for it!”

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.

Large language models (LLMs) offer powerful opportunities for automation, creativity, and knowledge access. With 800 million people using ChatGPT alone, LLMs are drastically changing the way we interact and operate as a society. No question, this is one of the biggest disruptions of our lifetime; but is it a good one?

We’re LIVE from DC Tech Connect, Batten Institute’s flagship tech event that brings students and alumni together to talk about digital tools and career paths. In this episode Yael and Mike explore LLMs with Dr. Ronnie Chatterji, OpenAI’s first Chief Economist. Chatterji is also the Mark Burgess & Lisa Benson-Burgess Distinguished Professor of Business and Public Policy at Duke University and has held previous roles with the Biden and Obama Administrations.

Check out Good Disruption Episode 28 at https://gooddisruption.podbean.com, or your favorite podcast platform.

You may not associate innovation with the library but look closer and you will find a quiet but relentless innovator. Working in libraries for 30+ years, I have witnessed libraries nestled within academic institutions not only survive but thrive. The work of librarians has drastically changed requiring us to find new ways to connect with faculty, students, and staff in a digital age. Innovation is a must for library survival.

Staying Relevant in a Fast-Changing World

The perception of the library is very traditional with books on shelves, quiet spaces, and yes, shushing librarians. Yet, librarians have been digitizing for 20 years. In 1996 the first library workshop I attended was at the UVA Alderman (now Shannon) Library. The library was leading the way on social science data services and the workshop covered how to use the Perl programming language (anyone remember Perl?) to place social science data online.

This was an early indication of what libraries would need to do to stay relevant. Over the next 30 years, the library transitioned from a majority print collection to an online collection that is available from anywhere in the world, 24 hours a day, seven days a week. Working closely with our IT friends, we have found creative ways to provide access to proprietary content online as seamlessly as possible for students, staff, and faculty. We have pushed and cajoled often reluctant publishers to provide their resources to academic communities and blazed the trail in negotiating licenses that would work both for us and for the publishers.

Our biggest successes have been with news and scholarly articles and data, though data still provides unique challenges for us. Publishers are reluctant to make their books available online to academic communities, as they would prefer us to continue circulating a single print copy to guard their content. Even here we are making strides, though, with collaborations between publishers and libraries bringing e-book packages to the academic market.

Our services moved online too. Community members can now chat, email, or meet with us online and regularly do so. I was meeting with students online long before COVID. When students went home to continue their studies online during COVID, I was ready and worked my schedule around zoom meetings with students in Europe, Asia, and Africa.

What about AI? We use Library resources such as Gartner and GlobalData Disruptor to follow the latest developments and forecasting and share those with the Darden community. Faculty regularly ask for our help in obtaining, when possible, large datasets. We are also helping people review AI generated citation lists to separate the real from the AI hallucinatory.

These are just a few examples of library innovation. In fact, libraries have led the way on many other initiatives such as open access, accessibility for individuals with disabilities, and information and data literacy. This has allowed us to connect with the larger goals and priorities of our colleges and universities.

The Secret Sauce to Survival

Why have academic libraries survived the waves of digitization?

Adding value. Long story short, the library saves universities and colleges millions of dollars every year. One example is the economist.com. If we paid full freight for a subscription for every student, faculty and staff member at Darden, it would cost $370,000 per year. We pay a fraction of that. Now multiply that by all the titles to which we provide access, and you get a sense of the scale at which we operate. And we add value to those collections by sharing our expertise via consultations, workshops, classroom visits, and research guides.

Cultivating close relationships with faculty, students, and staff. Library culture is a sharing culture and that culture builds close relationships. These relationships not only bring the library’s collections and services to life for people; it also keeps library staff aware of the latest trends in scholarship and learning. In a very real way, too, moving resources online has untethered us from a physical reference desk and allowed us to meet faculty and students where they are, online or in the classroom. The library community also shares ideas and, when possible, resources. This, along with innovative use of technology, has led to major improvements in interlibrary loans and consortiums that help us provide access to some of the larger publishing packages upon which we rely.

Balancing flexibility with tradition. Library staff have exhibited a healthy push and pull when it comes to technology, pushing to incorporate the latest technologies into our work while pulling to retain the traditions of our profession. Our long-term view of collections and services keeps us grounded and ensures that we collect and preserve traditional resources. For example, libraries continue to collect materials only available in print, which is the case for many foreign materials as well as many materials in the humanities and social sciences. We also keep in mind tomorrow’s faculty and students as well as today’s, constantly planning and preserving for the future.

Nurturing Creativity. I have found the people who work in libraries to be very creative, and library culture often encourages this creativity to flourish. I have worked alongside poets, authors, scholars, classically trained musicians and the list goes on. For example, Darden Camp Library has the benefit of a trained designer. They bring their creativity to work and push the library in new directions.

Riding the Winds of Tradition. Corporations spend millions of dollars on making their coffee shops look like traditional libraries. In surveys, people generally have positive feelings about the library. There is something comforting, I think, in the library as a constant, steady presence during chaotic and fast-changing times. As any Chief Marketing Officer will tell you, obtaining a market identity is extremely difficult, especially in this oversaturated marketing era. While a blessing, it also can be a burden, and the ongoing challenge is to ensure people are aware of both the traditional library and its modern digital counterpart. The library brand must evolve while honoring its decades-old tradition.

The next time you are in a quiet library surrounded by shelves of books, think of a swan. Beautiful and serene on the surface but paddling like crazy below the surface (or online).

The Batten Institute @ UVA Darden Substack is a space to share bold ideas and ask critical questions about what drives us forward. Our Batten community is broad and deep and our community members – faculty, staff, alumni and students – are invited to share their stories, ideas, and reflections on topics of interest to founders, innovators, and leaders everywhere.