| Episode | Status |
|---|---|
Spenser Skates has spent more than a decade building Amplitude from a YC startup into a public company, and in that time, he's had to reinvent himself just as much as the product.Joining the Light...
Spencer Skates, CEO of Amplitude, shares how his public company transformed from AI skepticism to building AI-native analytics products in under a year. He details the organizational challenges of pivoting a 200-person engineering org, the importance of bottom-up adoption over top-down mandates, and why incumbents struggle with AI transformation. Key insights include running an 'AI week' to train the entire team, doing two major reorganizations, and the fundamental difference between building SaaS (customer-driven) versus AI products (technology-first).
Amplitude was initially skeptical of AI from 2022-2023, with engineers frustrated by AI hype versus actual capabilities. The turning point came in October 2024 when they hired new AI-focused leaders and acquired Command AI. Spencer explains why engineers were rightfully skeptical of 'grifting' and how the jagged capabilities of early models made it hard to see practical applications.
Spencer describes running an 'AI Week' in June 2024 to train the entire 200-person product/engineering org. The key was getting leaders to use the technology first, then demonstrating capabilities (like live-coding dark mode) to show what's possible. This bottom-up approach led to all current AI products including their MCP server and AI visibility tool.
Spencer contrasts traditional SaaS development (ask customers what they want, build it) with AI product development which requires technology-first understanding. Because AI capabilities are 'jagged' (exceptional at some things, terrible at others), you can't just ask customers what they want - you need to understand model capabilities first then map to customer problems.
Spencer argues that AI adoption is uniquely top-down (unlike typical tech adoption) because Sam Altman is 'the best salesperson of this generation.' Executives and investors bought into the vision before capabilities caught up, creating frustration among engineers who saw the gap between promises and reality. This explains why many companies struggle with AI transformation.
Amplitude did two major reorganizations of their 200-person product/engineering org in 2024, moving out leaders suited for SaaS but not AI-native development. They acquired multiple YC companies (Command AI, Craftful, Inari, June) and blended those founders with longtime Amplitude employees. The key difference: AI-native builders understand technology capabilities first, while SaaS veterans are customer-driven.
Engineer Liu Zhang built AI visibility as a free product during AI week (he was planning to leave to start a company). The launch doubled new signups to Amplitude's free plan every week. Spencer argues AI visibility is a feature not a company - it's too easy to build and will commoditize quickly, but works great as lead generation for a real analytics business.
Amplitude didn't throw out their roadmap - they created four priorities: rebuild Amplitude to be AI-native, make it easier to use, bring other products to competitive parity, and serve marketers. AI features like 'Ask AI' (chat interface) make the existing product easier to use rather than replacing it. A dedicated AI team works on new products while the main team continues SaaS development.
Before Amplitude, Spencer and his cofounder built Sonolight, an early Siri-like voice recognition product at YC Winter 2012. Despite great press and demo day buzz, the tech wasn't good enough. They pivoted to analytics in June 2012 because they'd built their own analytics for Sonolight and other companies wanted it. Analytics was crowded but offered deterministic problems (right answers) vs probabilistic AI problems.
Spencer learned B2B sales by finding Mitch Morando, a sales exec turned coach, who would come in weekly and challenge him on customer pain points. The key insight: you can't learn sales from books - you need to do it and get coaching, like learning a sport. Be crystal clear on what you need to learn, then find mentors who can coach you through doing it.
The hardest transition for Spencer was unlearning 'leading from the front' on every problem. As a founder, you run to the hardest problem and solve it yourself. As a large company CEO with 800 people, you can't be in the weeds everywhere - you become the person you used to hate, judging others' work instead of doing it yourself. You must embrace hierarchy and be judicious with time.
Spencer's key advice: get crystal clear on WHY you're starting a company before beginning. Every successful founder in 'Founders at Work' reached a point where rationally they should quit, but didn't. The filtering criteria is intrinsic motivation - if you're doing it for recognition or money, you won't last through the 1-2 year rational quitting point. Dedicate yourself to a mission greater than yourself.
What Founders Have To Unlearn To Become Great CEOs
Ask me anything about this podcast episode...
Try asking: