Greg LeBlanc:
Welcome to Siloed. I'm Greg LeBlanc, and today I’m joined by Jeff Bussgang. Jeff is a Senior Lecturer at Harvard Business School and also the General Partner and founder of Flybridge Capital.
Now Jeff, I don’t know which is your main job—are you a lecturer and VC on the side, or a VC who happens to lecture?
Jeff Bussgang:
Honestly, I'm not sure either. I’d say I’m mostly a venture capitalist who lectures on the side.
Greg LeBlanc:
Well, that might be true—except you’ve written over 80 cases for Harvard and authored several books. There’s Mastering the VC Game, Entering StartUpLand, which is a career guide, and now your latest: The Experimentation Machine. Welcome, Jeff.
Jeff Bussgang:
Thanks, Greg. Great to be here.
Greg LeBlanc:
You kick off the new book by saying it’s about “timeless methods and timely tools.” That struck a chord. In both founding and investing, some things stay the same—timeless truths—while the tools are constantly evolving. Founders need to stay on top of those without losing sight of the core principles.
Another point I loved is your idea—building on Steve Blank—that a startup is a lab, and the CEO is the “Chief Experiment Officer.” I’ve been using that term too—I thought I came up with it!
So here’s my question: You argue that launching a company can be approached scientifically. But people often romanticize founders as visionaries or artists with unique charisma. So, how much of building a company is science, and how much is art?
Jeff Bussgang:
It’s both. Startups are highly non-deterministic—there’s no guaranteed formula. But there are timeless methods that improve your odds of success. You can’t guarantee outcomes like in engineering, but you can use structured processes to make better decisions.
And then, of course, there are timely tools. These days, new technologies are dropping not just every year, but every week, sometimes daily. Just recently, OpenAI released tools around deep research and agentic capabilities that some say are the most significant since ChatGPT launched. That’s just one company—every week, we see new breakthroughs.
So the challenge is: how do founders adopt these tools to become what I call “10x Founders”—like the mythical 10x engineers—but still adhere to those timeless principles?
Greg LeBlanc:
You wrote back in 2012 that every startup needed a machine learning PhD on the founding team. Was that tongue-in-cheek or something you truly believed? Today, with generative AI, it seems like the tools are accessible to everyone.
Jeff Bussgang:
Back then, it was true—machine learning expertise was a bottleneck. Today, it’s all about accessibility. Tools are abstracted to the point where the hottest programming language is now English.
We have no-code, low-code, and now AI-enhanced platforms that let anyone build powerful products. So instead of needing PhDs, we now need strategic thinkers who can apply these tools effectively. The experimentation machine is still key—it’s just powered differently.
Greg LeBlanc:
I usually teach strategy for mature companies. But when I shift to startups, it’s clear they don’t have answers—they’re in the discovery phase. How can first-time founders succeed in this process? And what’s the role of the venture capitalist—are they the coach helping prioritize which experiments to run?
Jeff Bussgang:
Absolutely. Entrepreneurship can be taught. I’ve been teaching at Harvard for 15 years—2,500 students have taken my course, and many had no startup background.
I’ve seen amazing success stories from people who came from big companies like GM and went on to build startups using the tools and structured processes we teach—customer discovery, prototyping, and hypothesis testing.
That search process is real. Chris Dixon calls it “idea maze” navigation. It’s not deterministic, but you can improve the odds with a scientific approach.
At the same time, some founders just have a flash of insight—often from their own domain experience or user pain—and build something great that way too. There’s no one path to success.
And yes, VCs should be coaches. When we invest, we’re trying to help founders navigate that maze—help prioritize, make sequencing decisions, and think strategically.
Greg LeBlanc:
You say strategy for a startup is all about test selection—figuring out which experiments will reduce the most uncertainty. That seems like a really powerful framing. How do you pick which experiments to run?
Jeff Bussgang:
That’s the billion-dollar question. Founders must choose experiments that test the most controversial assumptions and have the potential to unlock big valuation inflection points.
Take C16 Biosciences—a startup growing synthetic palm oil. They didn’t need to validate demand. Food companies want sustainable palm oil. The key was testing whether they could manufacture it at scale and hit price points. So they focused on experiments in the lab, not on customer discovery.
Every startup has a different key risk—your job is to identify it and test it first.
Greg LeBlanc:
So, no one-size-fits-all formula. But in general, you recommend focusing on the customer first—identifying a "hair-on-fire" problem, right?
Jeff Bussgang:
Exactly. Especially if you’re not in deep tech. Start with “the who.” Find the customer with a burning need. If you solve that, the rest follows.
Now, if you’re in a tech-first company—like synthetic biology or blockchain, you might start with “the what” and then figure out “the who.”
But often, engineers get excited by the tech and skip the problem phase. That’s dangerous. Smart founders often jump straight to solutions, skipping over customer discovery. And then they spend years building something no one needs.
Greg LeBlanc:
Speed is another big choice variable. You talk about scaling—some companies go too fast, others too slow. How should founders think about speed? Especially when technical debt becomes a tradeoff.
Jeff Bussgang:
Right—zero technical debt probably means you’re moving too slowly. But scaling too fast can kill a company. So it depends.
I outline several criteria in the book to decide if you’re ready to “blitzscale,” as Reid Hoffman puts it. These include strong product-market fit, capital availability, competitive pressure, and network effects.
More often than not, I advise caution. Most startups fail from premature scaling, not from going too slow.
Enterprise markets allow more time. But in consumer tech or AI, there might be a land grab—you may need to move fast or risk being left behind.
Greg LeBlanc:
So is it the VC’s job to adjust that pace? Maybe not “whipping the buggy,” but at least guiding with capital as the gas pedal?
Jeff Bussgang:
Fair. VCs can guide pacing decisions by asking the right questions: Do we have product-market fit? Are we ready to scale sales? Is the market pulling us?
We had a board meeting recently where we discussed this exact thing—how much to invest now versus later. Founders set the vision, but we help calibrate investment against risk.
Greg LeBlanc:
With generative AI, the ROI on experiments has exploded. People talk about the “one-person billion-dollar company” running agents. Is that hype or real?
Jeff Bussgang:
Sam Altman coined that phrase. I’m not sure we’ll see a one-person unicorn just yet—but a 25- or 50-person billion-dollar company? That’s already happening.
We’re seeing startups using AI agents for customer success, sales support, and engineering productivity. Not replacing people entirely, but making them dramatically more effective.
So yes, we may be entering a phase where the largest companies have fewer employees but more agents.
Greg LeBlanc:
But doesn’t that mean more competition? If everyone has the same tools—same agents, same LLMs—won’t the hit rate go down?
Jeff Bussgang:
Maybe—but the upside is also bigger. When my company went public in 1996, a billion-dollar market cap was huge. Now, $10 billion is the new $1 billion.
We’re not just attacking tech budgets—we’re attacking staffing budgets. Legal, customer support, and marketing—all are being disrupted by AI. So the outcomes could be even greater.
It’s more competitive, but the value creation potential is enormous.
Jeff Bussgang:
We're not talking about OpenAI or large foundation models here. These are specific applications in specific industries. So, how do these startups create moats? I mean, if 100 companies are all trying to automate discovery, what’s the differentiation?
Well, at Flybridge, we talk about three key sources of differentiation for AI application companies:
1. Proprietary Data Sets
For example, take Noticera—a legal AI company. They ingest multi-hundred-page legal agreements that law firms keep in their archives. These aren't publicly available, and OpenAI doesn't have access to them. By combining these with public data, they’ve built a rich database to help with contract negotiation in private credit financing.
2. System of Record and Last-Mile Workflow
This means becoming the environment where users spend their entire day. Think of AllSpice, a GitHub for hardware engineers—it’s both a collaborative cloud and an AI co-pilot. Hardware engineers live in it daily, just like software engineers live in GitHub.
3. Human-AI Interface
This is a really exciting frontier. The interface matters—whether it’s voice, chat, or something else. For instance, I created a voice clone of myself with D-ID. You could actually “interview” my AI clone—might’ve saved me from missing a ski day!
Greg (Interviewer):
That’s fascinating. But don’t incumbents like Salesforce or Workday already have the customer base and proprietary data? Doesn’t that give them an advantage?
Jeff Bussgang:
It’s a classic story. Every tech shift, incumbents claim they’ve got it all covered. I remember Steve Ballmer doing this with Microsoft 20 years ago, trying to steer VCs away from "his" markets.
Take MongoDB as an example. We led their Series A 15 years ago, betting on the shift to web apps and NoSQL databases. Despite Oracle's dominance, MongoDB carved out a huge space. Today, they’re worth over $20 billion. So yes, incumbents will win too, but so will AI-native startups finding those white spaces.
Greg:
You also talked about channel partners being both an opportunity and a risk. What’s your guidance for startups?
Jeff Bussgang:
Early on, focus on maximizing learning, not metrics. Channel partners can block direct access to customers, which slows your learning. Avoid gatekeepers. Later, once you deeply understand your customer, you can explore channel partnerships. But the mantra should be: Stay direct, maximize learning.
Greg:
But how do you measure learning? Most metrics like NPS or churn just show where you are, not what you’ve learned.
Jeff Bussgang:
That’s true—and it's a great question. I use a rubric called HUNCH for product-market fit:
H: Hair-on-fire value prop
U: Usage
N: Net Promoter Score (NPS)
C: Churn
H: High LTV-to-CAC
But as you said, these track results, not learning. The real signal comes from what we call the "earned secret"—those non-obvious, nuanced customer insights. They don’t show up in metrics, but they’re what separates average from great.
Greg:
So, pattern recognition helps, but startups are about doing things that have never been done. Can machine learning predict success in a venture?
Jeff Bussgang:
Venture is inherently non-deterministic. I can teach methods and frameworks, but founders have to apply them. Like Clay Christensen told Andy Grove: “I can share theories and case studies, but you have to decide how to apply them.” It’s why the case method is so effective for teaching entrepreneurship.
Generative AI's Impact on Founders
Greg:
Let’s shift to timely tools—how is generative AI changing the founder’s day-to-day?
Jeff Bussgang:
The first thing I have my students do is treat ChatGPT like a co-founder. Feed it everything—resumes, essays, past projects—and push it to ideate with you. It’s like having the smartest brainstorming partner ever.
Tools like NotebookLM or ChatGPT help you prototype experiments: test your value prop, identify your ideal customer profile (ICP), pressure test hypotheses. You can even throw in your sales call transcripts and ask how to overcome objections.
And it's not just about ideas—it’s also about building. My partner Jesse built a new app using Replit just by chatting on his phone. These tools are game changers.
Greg:
So it’s not just engineers who benefit—everyone from product to marketing can now scale with AI?
Jeff Bussgang:
Exactly. Product managers can auto-generate PRDs. Solo founders can use Spark Rockets to build landing pages. Customer service agents, sales scripts, investor updates—it’s all automatable now.
Greg:
I worked with a company recently. We uploaded all their public docs and bios into GPT. At the end of five days of workshops, we compared our findings to what GPT had generated on day one, and they were pretty close.
Jeff Bussgang:
That’s amazing. And yeah, it forces you to ask: what value are we adding? But the role of consultants and VCs is evolving, not disappearing.
The Evolving Role of the VC and the Founder
Greg:
So how does this change the job of the VC?
Jeff Bussgang:
We're already using AI to identify signals, automate outreach, manage CRM, and digest due diligence. I upload notes into NotebookLM so my partners can “chat” with the deal memo before the Monday partner meeting.
Greg:
Does this all make being a founder more accessible?
Jeff Bussgang:
Yes, I think so. Tools lower the barrier to entry. Non-technical founders can now build, test, and iterate on their own. Sam Altman talks about this in his New Year’s blog post—this explosion of creativity we’re seeing is real.
Greg:
So then, who do you want on your founding team now? What skills matter?
Jeff Bussgang:
I opened my book with this line: “AI won’t replace founders, but founders who use AI will replace those who don’t.” The same goes for employees. I think we’ll see a future where candidates show up with a “team” of AI agents they’ve trained to work alongside them—sales bots, data agents, whatever.
The bar for being “AI-native” is getting higher.
Greg:
So, these 10x or 100x founders—what makes them different now?
Jeff Bussgang:
They have an experimentation mindset. When experiments were expensive, prioritization was key. Now, experimentation is cheap. So your edge comes from being fast, iterative, and bold. You don't need perfect answers—you just need to keep learning.
Greg LaBlanc: So Jeff, are we now entering a world where the skill of identifying which experiments to run is more important than being able to run them?
Jeff Bussgang: Absolutely. Even though experimentation has become dramatically cheaper and faster, judgment and strategy will always hold timeless value, along with creativity. If you look at companies like Booking.com, they run thousands of experiments daily, testing everything from wording to pricing on their site. That level of instrumentation across the company is becoming more common.
But even in such environments, companies still need to make strategic decisions about what to measure and why. What are the true north metrics? When should you rely on data, and when is it time to trust your instincts?
Greg LaBlanc: You talk in the book about this tension between being data-driven and being hunch-driven.
Jeff Bussgang: Yes. And Fred Wilson of Union Square Ventures made an observation I liked: In the early days, founders should lean more on their hunches. Once you hit product-market fit, you become data-driven. That initial stage of prioritization—deciding which scarce resources to allocate where—that’s where human judgment is critical.
Greg LaBlanc: You've been teaching entrepreneurship and venture capital for a long time. Do you think these should remain separate courses? Or should we just be teaching all business students these skills, like experimentation, sequencing, and strategic inquiry?
Jeff Bussgang: I love that question. I’ve been surprised—but maybe I shouldn’t be—by how well executives at large companies respond to these concepts. My main audience is founders and entrepreneurs, but every company is trying to be more innovative now—nonprofits, governments, you name it.
My HBS colleague Mitch Weiss talks about the “one-mayor city hall” where a single civic leader can use AI to make decisions, write memos, pass ordinances—because they can access all the data and insights they need. So yes, this mindset is becoming universal.
Greg LaBlanc: And what about applying this mindset to personal life? Students often ask how these concepts apply to their careers—treating life itself like a startup, making strategic bets, and running experiments.
Jeff Bussgang: I’ve seen that mindset applied in meaningful, productive ways. Over the years, I’ve taught over 2,200 students, and many of them use entrepreneurial thinking to shape richer, more intentional lives.
Greg LaBlanc: So, people might one day use ChatGPT to ask who they should marry?
Jeff Bussgang: [Laughs] I’ll stay out of the AI girlfriends and boyfriends debate. But yes—this mindset of constant iteration, learning, and experimentation is powerful, not just in startups, but in life.
Greg LaBlanc: Jeff, thanks so much for joining me. The book is The Experimentation Machine: Finding Product-Market Fit in the Age of AI. It’s all about timeless methods and timely tools. Check it out.
Jeff Bussgang: Thanks, Greg. Great to be here.