Before You Read

Last weekend, AlgoVerde CEO Vladimir Jacimovic gave the commencement address to the Harvard Business Analytics Program. The talk was titled When AI Agents Change the Rules, Who Wins? The full text is below. Keep reading!

When AI Agents Change the Rules, Who Wins?

I’m privileged and honored to be here — and it’s a little funny how I keep getting invited to this program.

I was here at the very beginning. Years ago, when Karim Lakhani and David Parks were first sketching out the need for a program like this — a place where working professionals could learn about the intersection management and data science disciplines which did not exist in 2017: how to actually use analytics and AI in your job. That need was just becoming obvious, and this program was built to meet it. Since then, something like twelve, maybe fourteen hundred of you have come through it.

And here’s what makes it funny. They tend to invite me at the moments when something changes. The first time, because a new need was being born. And now — because the world has changed again, just as dramatically.

The good news and the bad news

There’s a concept in evolutionary biology called punctuated equilibrium. The idea is simple: for long stretches, the world is stable — things change slowly, at the margins. And then something happens, and very quickly the whole order reshuffles. New winners, new losers, a new landscape.

I’m here to give you the bad news and the good news. The bad news: we have just hit one of those moments. The equilibrium has been punctured. You can debate it, you can be in denial about it — but I’ll show you why I’m certain the line has been crossed. The good news — and I promise you a happy ending — is that you, specifically the people in this room, are unusually well equipped to thrive in what comes next. If you follow a few simple principles.

So what punctured the equilibrium? It isn’t AI in general. It isn’t even generative AI. It’s a new layer that now sits on top of them — agentic AI.

The three brains

Think of it as a brain we’ve been building one piece at a time. First we built the left brain — the analytical one. Logic, math, prediction. That was predictive AI, and when we created this program, that was the whole game: I want to teach you to do predictive analytics. Then we added the right brain — the creative one, that can brainstorm and synthesize and imagine. That was generative AI.

But here is the thing about a left brain and a right brain sitting side by side: on their own, they don’t actually do anything on their own. You can know the answer, you can dream up the possibilities, and still never get out of the chair. What was missing is the part on top — the executive function that sets a goal, makes a plan, takes the first step, checks the result, and adjusts. That is agentic AI. That’s the part that finally gets up and does the work.

A quick poll

Let’s do a quick poll. How many of you have heard about AI agents? How many of you have a CIO who has announced your company now has three thousand agents? And how many of you have seen your company make real, serious money from those agents?

That gap — everyone’s heard of them, half are “deploying” them, almost nobody is winning with them — that gap exists for one reason: they aren’t really understood yet. So let me try to fix that.

Why this is a real break — two breakthroughs

Agents didn’t appear yesterday. What happened is that the models powering them crossed two thresholds in quick succession.

One: they learned to loop and reason. The old way, you submitted a prompt and got back an answer — one shot. Now the model actually thinks before it answers. An agent produces an answer, then asks itself, “is that actually right?” — and tries again. And again. It iterates its way to a good answer, the way a person does. It’s far, far better than it was even a year ago.

Two — and this was the big one, around the end of last year: they learned to create other agents. I was mightily proud of myself when I built my first agent. Then my first three. Now those three spin up dozens more that do the work and report back. They propagate. You can raise your own little army.

What all of that unlocked: software became almost free

Now, that’s a lot of Silicon Valley techie-talk. Here’s why it matters. All of this new capability allowed one thing to be done exceptionally well: developing software.

I’ve spent my whole career around software companies, and for that entire time, building software was expensive and slow. You needed a team — the famous scrum team of eight: front-end developers, back-end developers, a QA engineer, a designer, a product manager. It cost real time and real money, and that’s exactly why the great franchises got built — Microsoft, Oracle, Salesforce.

Now that agents write the software, something that was very expensive is becoming almost free. And once software is nearly free, you can start treating it as disposable — almost like fast fashion. I have a problem; I have an agent build a small program to solve it; I use it; I throw it away. That was not remotely plausible nine months ago. And because almost any problem that can be solved with software now can be — software can’t change your toilet bowl, but it can absolutely tell you how — software has become a universal conduit for solving problems. That is what punctured the equilibrium.

And one confession before we go further: this whole talk has an expiration date. Everything I tell you today, by November there will be a new reality on top of it. Keep that in mind — it’s part of the point.

Three examples from the front lines

Fair enough, you say — you live in San Francisco, of course you’re excited. What does it mean for me? Let me give you three quick views from the front lines.

One — the speed of software development itself

Take Anthropic, one of the leaders here. In a recent stretch they shipped 74 product releases in 52 days — and these are new things, not little tweaks. How? Because roughly 80% of the code Anthropic ships is now written by their own AI, and their engineers are developing something like eight times as much code per day as they were two years ago. The tool builds the tool, which builds the next tool. That’s the flywheel.

Two — rethinking the whole organization

If the models are this good and I can field an army of them, I can rethink how my company is built. For two thousand years hierarchy existed just to route information — for both militaries and private enterprise alike. Now the knowledge base does that: every meeting, document, and customer call goes into one repository, and agents mine it day and night, building a live model of the whole company. That replaces the middle layer. People move to the edge, where judgment matters. And the agents? They don’t complain, they work 24/7, you can’t even get annoyed — they’re not human.

Three — from my own backyard

One of my companies, AlgoVerde, does exactly this — and it’s public knowledge, so I can name it: Nissan is a major customer, and we’re reinventing how automakers design cars. When you build a vehicle, the hardest part comes first: the concept phase. You have to find the white space in the market, pin down the target customer, and decide which features — features that have to be cost-effective and still feel fresh three or four years from now — go into a car you’ll then sell hundreds of thousands of. That work has traditionally taken twelve to eighteen months.

Here’s how the agents do it. We stand up agents that behave like synthetic personas of real customer segments — so you can “interview” the market in an afternoon instead of fielding studies for months. And we stand up another set of agents that orchestrate the workflow — running the analysis, pressure-testing the feature set, assembling the story for management. Together they’ve taken that concept phase from eighteen months down to about three or four. It’s in production, delivering value — and, not coincidentally, Nissan has its mojo back.

So what do you do Monday morning?

That’s the techie tour. Now the only question that matters: what do you do Monday morning when you walk into the office? And here is where the good news comes in.

Since the very first cohort walked into this program, this group has had two distinguishing features. The first is an insatiable desire to learn — not just ambition, not just wanting the next title, but a genuine drive to keep bettering yourselves. The second is this community — the peers and friends you built here who help each other. For a largely hybrid program, that always amazed me. Hold onto both. They’re about to be your biggest advantages.

Because this shift is going to land on you in three places at once. But first you have to become an orchestrator.

First, on you personally. You’ll have to run this agentic stuff yourself — and that means more than buying a Claude subscription and using it for homework. How do you actually get better? How do you stay informed? That alone is nearly a full-time job. Let me tell you how fast this moves: at Christmas, Google’s Gemini was the only name in town. By the third week of January, Anthropic had leapfrogged it, and we’ve all spent the last four months sprinting to keep up — until the next thing happens. You have to make that investment anyway.

Second, on your teams. Most of you lead people, and those people are anxious. Some days I feel less like a CEO and more like a therapist for my engineers, because we use agents for everything. “Will we even have jobs in 24 months?” Yes — and a better job than you have today. But you have to lead them there, and teach them how to thrive in it.

And third — the hardest one — managing up. For a century, companies have been built like a Roman legion — the Alfred Sloan structure: a CEO with six or eight reports, each of them with six or eight reports, all the way down. We built it that way to move orders and information cleanly. Well, the agentic world is a wrecking ball to that paradigm. And you’ll find yourself in a genuinely awkward spot: if you want to adopt this, you’ll be pushing against a structure that has been stable for a hundred years. That’s the very definition of punctuated equilibrium — a long, stable order about to reshuffle.

Learning to unlearn

Now, you might be thinking: with all these agents doing the work, life finally gets easier. I’m sorry to tell you — no. It gets harder. When you can suddenly do this much, the ambition expands to fill it. I’ll prove it to you with my own house: I have three children, and they bring their laptops to the dinner table — not to ignore us, but because they’re running their agents while we eat. That’s the new normal you’re walking into.

But here is why I’m optimistic about you. Very few people on earth will know how to navigate this new universe — and you are unusually well prepared for it. Not because of any one technique you learned here. Nobody’s edge in this era is going to be that they remember how to do K-means clustering. Your edge is a way of thinking, and that insatiable desire to learn.

And there’s one last skill I’d add to it — and it’s the title I want to leave on the screen as you walk out. In a punctuated equilibrium, the people who thrive aren’t just the ones who learn the fastest. They’re the ones willing to unlearn — to put down the old habits, the old org chart, the old way the job used to work, and pick up something new. Punctuated equilibrium is the model that explains this moment. Learning to unlearn is how you survive it — and then thrive.

You’ve already proven, twice over, that you know how to do the hard part: you keep choosing to learn. So keep going.