How to Lead Your Startup Through AI Transformation: Lessons from Amplitude's CEO

Key Takeaways

• Perseverance over profit: The critical difference between successful founders and those who quit is pushing through the 1-2 year low point when logic says to give up
• Clarity on "why": Crystal clear understanding of what you're building and why you're building it anchors you through uncertainty and emotional pain
• Technology-first thinking for AI: Unlike traditional SaaS (which relies on customer feedback), AI product building requires deep understanding of what's technically possible first
• Organizational transformation requires hands-on leadership: Training your entire team to understand and use new technology, not just talking about strategy
• Self-selection and mental model shifts: The best team members naturally gravitate toward new opportunities when leadership demonstrates genuine conviction and models the behavior

The Founder's Critical Crossroads: Why Most Startups Fail at Year One

Building a startup is, quite frankly, one of the most emotionally painful experiences you can willingly put yourself through. When you're roughly one to two years into founding, you hit a specific moment—a rational inflection point—where the numbers tell you to quit. The market might be slower than expected. Your revenue might be plateauing. Your team might be questioning direction. Every logical analysis suggests walking away.

Yet here's what separates the founders who build billion-dollar companies from those who become cautionary tales: the successful ones don't quit at that moment, even when everything rationally suggests they should.

This isn't luck. This isn't some mystical founder gene you either have or don't have. The founders who survive this crucible share one thing in common: absolute clarity about why they're building in the first place.

Spenser Skates, CEO and co-founder of Amplitude (a leading analytics platform trusted by companies like DoorDash, Walmart, and Cursor), calls this "getting that top node really right." Think of it like a decision tree in computer science—the top node is your foundational reason for existing. Everything branches down from there. If that top node is fuzzy, vague, or externally motivated (I want to get rich, I want recognition, I want to prove something to someone), your ability to survive the valley of despair diminishes dramatically.

But if your top node is crystalline—"I'm building this because I believe it will fundamentally improve how people work" or "I see a problem that's been unsolved for a decade and I'm obsessed with solving it"—you have an anchor point. When your cofounder wants to quit. When your largest customer churns. When you realize you built the wrong product. You go back to that top node, and it sustains you.

The founders who fail often come with a hedged bet: "I'll do this startup, and if it works out great; if not, I'll go to grad school or get a job." This conditional commitment is, as Spenser puts it bluntly, "the freaking worst." Why? Because startups require you to sit with existential uncertainty for long periods. You can't be half-committed to that uncertainty. You'll burn out before you break through.

The SaaS Playbook That No Longer Works: Why AI Forces a Complete Mental Reset

For the last decade, B2B SaaS companies have operated from a proven playbook:

1. Talk to customers
2. Ask them what they want
3. Prioritize the list based on revenue potential
4. Build it
5. Ship it to customers
6. Repeat

This loop is extraordinarily powerful. It's how Amplitude itself scaled to become a market leader. It's the core competitive advantage of mature SaaS companies.

But here's the problem: this playbook doesn't work for AI products.

The fundamental issue is that customers don't yet know what's possible. If you ask your customers what AI capabilities they want from your product, they'll ask for a "faster horse"—incremental improvements on the existing interface. They literally cannot envision what becomes possible when you rebuild the entire experience around AI-first thinking.

This is where the mental model shift becomes critical. For AI products, you have to reverse the order:

1. Deeply understand what the technology can actually do (not what vendors claim it can do, but what it genuinely excels at and where it fails)
2. Map those technical capabilities to customer problems in ways customers can't yet articulate
3. Build from a technology-first perspective, not a customer-feedback-first perspective
4. Let customers experience it, then gather feedback and iterate

This is profoundly different from the traditional SaaS mindset, and it explains why so many established tech companies struggled with AI adoption. Their engineers—people who built incredibly successful products by listening to customers—suddenly found themselves in a world where customer listening was actively harmful to product development.

At Amplitude, the skepticism about AI ran deep initially. Cofounder Jeffrey saw "grifting"—hype and smoke with no real substance. And honestly, for much of 2023 and early 2024, he was right. The capabilities of large language models were jagged. They excelled at some tasks and were absolutely terrible at others. Telling an engineering team "you should do more AI" without understanding those distinctions felt insulting and misguided.

It wasn't until the second half of 2024 that Amplitude's leadership saw proof that AI could genuinely transform what analytics could do. They saw it in their own workflows when they started using tools like Cursor for software development. That observation—watching their own engineers become dramatically more productive with AI assistance—created the conviction necessary to transform the company.

The "AI Week" Moment: How to Get Your Organization to Actually Believe in Change

Here's a critical lesson that most large companies get wrong: talking about AI transformation is not transformation.

After Amplitude's leadership decided to get serious about AI, their first instinct was to tell the organization "we need an AI strategy." This produced the expected resistance. People were tired of hearing about AI. They were focused on shipping real products that made real money.

So instead of more talk, Spenser and the team did something radical: they had everyone actually use the technology.

They created "AI Week"—a dedicated sprint where the entire product, engineering, and design organization got hands-on with Cursor, AI models, and other cutting-edge tools. They didn't hold lectures. They didn't create frameworks. They had their VP of Product sit down and code a dark mode for Amplitude in front of the whole company.

That moment—watching leadership actually using these tools to create tangible results—changed everything. It wasn't a mandate from the top. It was proof that the technology worked. One of their product leaders saw this and thought, "If all the leaders are doing this, I better get on board." Another engineer, Leo Jang, who was planning to leave Amplitude to start his own AI company, decided to stay and dive deep into AI visibility instead. They gave him the resources and freedom to pursue it.

What happened next? His AI visibility product doubled new sign-ups to Amplitude. Every week, they were getting twice as many new customers as they had before the feature launched.

This is the real playbook for organizational transformation:

1. Don't mandate change; demonstrate it
2. Let people experience the technology themselves before asking them to commit
3. Create space for self-selection; people who are genuinely excited about the new direction will naturally gravitate toward it
4. Give those people resources and autonomy to go deep
5. Celebrate the wins publicly so others see that this isn't just executive posturing

Hiring and Structural Changes: Bringing in AI-Native Thinking

However, organizational transformation isn't just about getting existing people excited. Sometimes, you need to fundamentally change who's in the room.

Amplitude made two major organizational restructurings in 2024. Some leaders and managers who were exceptional in the pure-SaaS era weren't the right fit for an AI-native future. They understood product development in the SaaS world, but they didn't have the mental models to navigate the uncertainty and novelty of AI product building.

Rather than forcing these people into new frameworks, Amplitude made the difficult decision to move some of them out and brought in people with deep AI expertise. This included hiring Wade Chambers, a Silicon Valley legend with years of experience deploying cutting-edge AI models in production. They also acquired entire teams—Command AI, Craftable, Manari, and June—bringing experienced founders into the organization.

This wasn't about replacing people who "didn't get it"; it was about recognizing that different eras require different skill sets. It's possible (even common) for someone to be exceptional in one era and less suited for the next. The key is recognizing that early and making structural changes rather than trying to force a round peg into a square hole.

The combination of long-time Amplitude engineers (who understood the analytics problem deeply) with AI-native founders and engineers (who instinctively thought in terms of what new models could enable) created something genuinely special. They had historical product knowledge married to cutting-edge technical capability.

The Mentor Model: How Great Founders Learn Skills They Don't Already Have

One of Spenser's most underrated insights comes from his experience learning B2B sales—something he had absolutely no background in.

Amplitude is fundamentally an enterprise product. You can't just convince one person to adopt it; you need to convince multiple stakeholders. This requires sophisticated sales motion, and Spenser knew nothing about it.

Rather than reading books about sales or taking online courses, he did something smarter: he found Mitch Mirando, an experienced sales executive who had coached other founders, and had him come in weekly to coach him directly.

Mitch would watch Spenser's sales calls and push back hard. When Spenser would say, "The customer wants better dashboards with charts," Mitch would respond, "Spencer, that's not a business pain. What's the pain?" Over time, this direct feedback loop, combined with repeated practice, fundamentally changed how Spenser thought about selling.

This is the critical insight: learning complex skills requires three things—clarity on what you need to learn, direct practice, and coaching from someone who's already mastered it. Reading and studying are supportive, but they're not sufficient.

For founders just starting out, this means:

1. Be crystal clear on what you need to learn right now (not everything, just what's blocking progress)
2. Find someone who's genuinely excellent at that skill (look for people with proven track records and evidence of success, not just credentials)
3. Get them to coach you directly, ideally in real-time or with short feedback loops
4. Be willing to look like you don't know what you're doing while you learn

The best mentors will be uncomfortable for you at first. They'll push back on your thinking. They'll point out flaws in your reasoning. But that discomfort is the signal that learning is happening.

The Paradox of Founder Mode at Scale: When Hands-On Leadership Becomes Impossible

Here's a tension that most startup advice doesn't adequately address: the skills that make you a great early-stage founder actively work against you as a large company CEO.

As a founder, your instinct is to run headfirst into the hardest problem. If there's a difficult piece of code, a tricky product design decision, a struggling customer relationship, or a team conflict, you jump in and lead from the front. You rally people around you by example.

But when you have 800 employees and you're running a $350 million revenue company, you can't jump into every hard problem. You can't code. You can't sit in every important customer call. You can't personally resolve every team conflict. If you try to do that, you become a bottleneck. Nothing scales.

So what happens? You become the thing you always mocked—a large company executive who delegates everything and spends all their time "judging other people's work." There's good reason this exists, though. It's not laziness or corporate bureaucracy. It's mathematics. There are simply more problems than hours in your day.

The transition from founder-mode problem-solver to large-company-executive-strategist is, by Spenser's own admission, "the hardest transition by far." It requires learning new skills:

• Strategic time allocation: Not saying yes to everything, being ruthless about what you focus on
• Delegation and trust: Genuinely empowering people to own problems rather than trying to oversee everything
• System and hierarchy: Recognizing that some level of organizational structure is necessary, even though it frustrates you
• Communication and alignment: Ensuring thousands of people understand the direction without you personally convincing each one

The wrinkle is that "founder mode" for a large company isn't about doing everything yourself; it's about having a very clear point of view on direction and being visibly committed to that direction. At Amplitude, this meant Spenser being visibly excited about AI, using the technology himself, experimenting publicly with it, and making it clear that this was a genuine priority (not something middle management should handle).

The "Cursor Moment" in Every Industry: Where Startups Win Against Incumbents

One of the most valuable frameworks Spenser articulates is the idea of the "cursor moment"—a point where new technology becomes so obviously superior that going back to the old way feels impossible.

Cursor was the first AI coding assistant that felt genuinely better than coding without AI. It wasn't incrementally better; it was transformatively better. Developers using Cursor couldn't imagine going back to coding the old way. This created a market opportunity for the entire analytics space.

There's going to be a similar "cursor moment" in analytics in the next one to two years. Analytics platforms will be rebuilt around AI conversations, where you can naturally ask questions and get answers, and suddenly the old way of building dashboards and queries will feel antiquated.

For founders thinking about which incumbent markets to attack, this is the key question to ask: Is this a category ripe for a cursor moment? Is there a large, expensive, legacy incumbents solution that could be fundamentally reimagined with AI?

Email is a massive opportunity (Google's email product is astonishingly poor). Workspace software has huge potential. Tech support is ripe for disruption. Marketing technology has been underserving customers for years.

The advantage you have as a startup isn't superior resources (incumbents have those). It's that you don't have to maintain backwards compatibility with millions of customers who are used to the old way. You can rebuild from first principles. You can bet everything on what AI makes possible rather than incrementally improving what already exists.

But here's the catch: building a category-defining product in the AI era requires deep understanding of both the technical possibilities and the actual customer problem. Too many AI startups are building "visibility" products (which are table stakes now, not defensible businesses). The real businesses downstream are those that solve actual business problems—content generation (like AirOps), security and compliance concerns, support automation, and so on.

The Emotional Reality of Startup Leadership: Why Conviction Matters More Than Comfort

This conversation with Spenser returns repeatedly to one theme: starting a company is emotionally brutal, and the only thing that sustains you through it is genuine conviction about what you're building.

Investors will pressure you to adopt AI. Board members will ask for your "AI strategy." The internet will mock you if you're doing something they think is outdated. Media narratives will shift underneath you. Customers will demand features you don't believe in. Your own team will question direction.

The only thing that carries you through all of that is knowing, at a deep level, why you're building what you're building. Not for recognition. Not for money. Not to prove something to someone. But because you genuinely believe you're solving an important problem and you're willing to sit with years of uncertainty to prove it.

For Spenser, that conviction about analytics has carried him through multiple market cycles, product pivots, and now a fundamental reimagining of the entire category. That conviction is why he was willing to spend a year getting Amplitude's organization onboarded and believing in AI, rather than shipping AI features to try to appease investors and customers.

This is the real founding lesson: Build clarity on your why. Get ruthlessly honest about whether you're willing to stick with it through the hard moments. And then be willing to learn whatever skills and technologies you need along the way to actually make it happen.

Conclusion

Scaling a startup through industry transformation requires three interconnected commitments: unwavering clarity about your purpose (to anchor you through the inevitable low points), ** willingness to fundamentally reshape your thinking and team** (as new technologies and markets emerge), and ** commitment to continuous learning** (by finding mentors, using new tools yourself, and letting your team experience innovation firsthand).

The AI era isn't going to reward companies that talk about transformation. It's going to reward companies that live it—whose leaders actually use the technology, whose organizational structures reflect new priorities, and whose teams have genuine conviction about what becomes possible.

If you're building a startup right now, the question isn't "Should we add AI to our roadmap?" The question is deeper: Are we clear about why we're building? Are we willing to sit with uncertainty long enough to find the genuinely transformative applications of these technologies? And do we have the right people and the right conviction to get there? Answer those three questions with honesty, and you've got a chance to build something remarkable.

---

Original source: https://youtu.be/t8co94HS6tY?si=UagjaTevoKvTjoUj

powered by osmu.app