Apr 3, 2023 · 35m · mad
A Conversation with Sarah Catanzaro, Amplify Partners
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the MAD Podcast at Data Driven NYC, Matt Turck interviews Sarah Catanzaro, General Partner at Amplify Partners, to discuss venture capital thesis design, data stack economics, and supporting technical founders. Sarah provides strategic perspectives on navigating generative AI hype, incumbent competition, developer tools, and the shift toward localized compute engines.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 12.6% of the talking time here. How this is scored →
Matt as informed peer 2.3 Guest teaching 3.8 Guest disagreement 1.2 Matt pushing back 0.7
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 9:45 Rejecting AI founder FOMO and championing second-mover advantage
Sarah directly rejects the prevailing market narrative that founders must start an AI company immediately or miss the wave, arguing instead that second movers will benefit from observing early failures.
Hardest push from Matt ▶ 10:51 Challenging startup viability against proactive incumbentsMatt presses Sarah on whether startups can build enduring standalone companies when big tech incumbents are leading AI development rather than acting as slow incumbents.
Biggest teaching moment ▶ 17:15 Breakdown of compute vs storage costs driving data architecture changesSarah educates on how the modern data stack's push to ELT relied on cheap storage, but escalating warehouse compute bills are forcing the market back toward ETL and edge transformations like DuckDB.
Matt holds his own ▶ 13:39 Demonstrating domain fluency in LLM Ops infrastructureMatt shows strong technical market knowledge by introducing the emerging 'LLM Ops' stack and prompting Sarah to evaluate where value will accrue across infrastructure versus application layers.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Sarah Catanzaro's Career Background and Path to VC | 0 | 2 | 0 | 0 | Sarah details her career path from studying insurgencies with symbolic systems to working at Palantir, Mattermark, and eventually venture capital. The host merely introduces the topic with no active technical contribution. | |
| Amplify Partners' Thesis and Investment Focus | 1 | 2 | 0 | 0 | Matt prompts Sarah to explain Amplify's firm thesis. Sarah explains their explicit focus on technical tools and infrastructure over generalist investing. | |
| Evaluating the Current AI Hype vs. Past Cycles | 2 | 4 | 2 | 1 | Matt prompts a comparison between the current AI wave and past cycles like 2017. Sarah disagrees with current market consensus and FOMO, pointing out potential second-mover advantages. | |
| Startups vs. Incumbents in the LLM Era | 3 | 5 | 2 | 2 | Matt challenges whether startups can withstand non-lazy big tech incumbents in AI. Sarah highlights how rapidly consensus shifted over a single month as incumbents proved surprisingly fast at integrating LLMs. | |
| Value Creation across LLM Applications and Infrastructure | 4 | 5 | 1 | 2 | Matt brings up LLM Ops and asks if traditional data infrastructure is losing momentum. Sarah educates on how high compute costs are driving the industry from ELT back to ETL and edge transformations. | |
| Deep Dive into DuckDB and MotherDuck | 3 | 5 | 2 | 0 | Matt asks for a deep dive into portfolio company MotherDuck and DuckDB. Sarah challenges industry consensus by stating 'big data is a lie' for the vast majority of organizations. | |
| Partnering with and Scaling Technical Founders | 3 | 4 | 1 | 1 | Matt asks about scaling deeply technical founders into company leaders. Sarah shares a story highlighting how deeply technical founders often lack basic mental models for enterprise sales organization. | |
| Continuous Technical Learning and 'Projects to Know' | 4 | 3 | 2 | 1 | Matt praises the technical depth of Sarah's newsletter 'Projects to Know'. Sarah jokingly suggests asking ChatGPT where to find it and details how she caps pitch meetings to preserve time for research. | |
| Audience Q&A: The Future of Edge AI and Latency | 1 | 4 | 1 | 0 | An audience member asks about Edge AI. Sarah reframes the edge value proposition away from privacy and toward latency, noting how slow existing LLM applications currently are. | |
| Audience Q&A: AI Explainability and Safety Metrics | 2 | 4 | 1 | 0 | Responding to an audience question about black box models, Sarah notes that research has shifted away from 2017-era explainability toward model evaluation and safety metrics. |
Statements from this episode (18)
Catanzaro: Technical tool startups face homogenous go-to-market challenges
“I think the challenges that technical tool builders face actually tend to be like fairly homogenous, whether it's thinking about like open source strategy, Thinking about like top down versus bottoms up execution or, you know, what that even means thinking abo…”
Catanzaro: AI technology is not a substitute for product fundamentals
“Applying AI is not a substitute for building great products. The kind of same principles of delivering value to users, of managing kind of their attention, all of those principles hold. You know, AI is just a technique. It's just an underlying technology that …”
Catanzaro: AI startups prioritize hype and attention over practical utility
“Instead of focusing on delivering concrete value to people and almost delivering like Mundane value to them They. They seem to be more focused on competing for mindshare. It almost feels like a lot of AI startups today kind of think about themselves as being e…”
Catanzaro: Second-mover AI startups can benefit from early failures
“I actually disagree with that. I think like we are learning so much literally day by day. I mean, GPT four was released yesterday. And there is possibly, you know, a second mover advantage. There will be lessons that we learn again from kind of seeing this fir…”
Catanzaro: Incumbents can move swiftly to integrate LLMs into applications
“I think the past month has demonstrated that that is not true, that, you know, a lot of existing companies can actually act pretty swiftly to integrate LLMs into their applications.”
Catanzaro: Prototyping LLM applications no longer requires a dedicated ML team
“It is so much easier to prototype LLM driven applications. You no longer need to hire an ML team before you can even determine, before you can even run an AB test or experiment to determine, like, if there, if this is something that users like.”
Catanzaro: AI startups must reimagine workflows rather than iterate existing tools
“And I think that that is the promise of LLMs and startups. It's not to build better versions of things that exist today, but rather to kind of reimagine some of the tools and systems that we use in an LLM centric way.”
Catanzaro: Value in AI will accrue at both application and infrastructure layers
“I don't think we'll see that value is accruing exclusively at the infralayer or the app layer. We haven't seen that in any other paradigm shift. Again, if you think about, you know, mobile, if you think about like social networks, et cetera, there was value at…”
Catanzaro: AI infrastructure adoption will surge after teams struggle without proper tools
“It feels like right now some of the value at the app layer can be captured immediately. Whereas the value at the infra layer will expose itself over time because so many of these companies who are, you know, trying to integrate LLMs into their existing product…”
Catanzaro expects current modern data stack tools to be replaced
“I expect the same to happen with regards to the modern data stack.”
Catanzaro: Cloud data warehouse transformation costs are surging due to compute
“Storage is cheap, but compute is not. And if you are preparing your data in a data warehouse, you're using compute, and your bill is going to get pretty high.”
Catanzaro: Most companies do not actually have big data
“Most companies do not in fact have like big data. Perhaps there are a few companies like in genomics or like astrophysics that like actually have, you know, Petabytes of data, but like most data sets are not that big.”
Catanzaro: Modern data tools are over-engineered for daily analyst workflows
“So many of the kind of data tools and systems that people use today, they're optimized for big data. They're optimized for running a point query across petabytes of data in milliseconds which is not the work that most data analysts and analytics engineers are …”
Catanzaro: There are few discernible patterns that predict founder success
“I, I've been investing for seven, eight years now, and it's something that I think about a lot, but frankly, like, I don't see that many patterns related to, you know, successful founders or those who don't succeed.”
Catanzaro: Technical founders shouldn't learn business functions from first principles
“I find myself constantly, like, reminding technical founders, including those within the Amplify portfolio, that, like, they don't need to learn things from first principles, or even from experimenting, like, They can just ask people who know how to do it, and…”
Catanzaro limits her schedule to three startup pitches per week
“I try not to see more than, like, two or three pitches a week.”