Aug 23, 2024 · 48m · sourcery
Computer Science is 'So Hot Right Now' | Grace Isford, Lux Capital · Sourcery with Molly O'Shea
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 'Sourcery,' host Molly O'Shea interviews Grace Isford, Partner at Lux Capital, to explore the rise of New York City as an artificial intelligence powerhouse, Lux's technical investment thesis, and practical strategies for enterprise AI deployment.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Molly holds 17.6% of the talking time here. How this is scored →
Molly as informed peer 3.9 Guest teaching 5.8 Guest disagreement 0.4 Molly pushing back 0.4
speaking balance: gold is Molly, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 38:39 Nuancing valuation drivers
Isford gently pushes back against Molly's broad generalization that multiples are primarily driven by team pedigree rather than revenue, citing rapid ARR acceleration across top portfolio companies.
Hardest push from Molly ▶ 30:51 Refusing the NYC relocation pitchAfter Isford delivers a multi-minute pitch on why NYC is surpassing SF in AI momentum, Molly directly rejects the idea of moving back to New York.
Biggest teaching moment ▶ 25:31 NYC AI ecosystem deep diveIsford delivers an authoritative breakdown of NYC's structural advantages in AI, citing academic lab feeders, Fortune 500 customer proximity, and portfolio migration statistics.
Molly holds their own ▶ 31:26 Framing AI market correction and acqui-hiresMolly demonstrates industry expertise by synthesizing the shift from mega-rounds to acqui-hires and questioning the sustainability of premium valuation multiples.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Molly as informed peer | Guest teaching | Guest disagreement | Molly pushing back | Why |
|---|---|---|---|---|---|---|
| Show Animated Opening Bumper Sequence | 4 | 4 | 0 | 0 | Molly opens the show highlighting Isford's background and achievements at Lux Capital and Canvas Ventures. The exchange is warm and collegiate as Isford explains how her time at Handshake inspired her focus on computational sciences. | |
| Research Methodology and Data Stack Evolution | 3 | 6 | 0 | 0 | Molly asks Isford to break down her specific research process and thesis creation. Isford walks through how asking foundational questions leads to expert interviews, recursive flywheels, and market maps that parallel early data stacks with current AI infrastructure. | |
| Global Deal Sourcing and Sakana AI Overview | 4 | 6 | 0 | 0 | Molly asks how Isford found and decided to invest in Sakana AI in Japan despite Lux's heavy NYC presence. Isford details the global distribution of AI talent and the specific background of co-founder David Ha. | |
| Sponsor Segment: Archer Aviation Electric Air Taxis | 3 | 5 | 0 | 0 | Following a sponsor read, Isford details Sakana's co-founding team and outlines Lux's deal sourcing strategies and check-writing range from seed to growth stage. | |
| Decoding Hugging Face's Monetization Model | 4 | 6 | 0 | 0 | Molly asks about Hugging Face's business model and how Lux conducts rigorous technical due diligence. Isford explains Hugging Face's freemium and enterprise model and highlights the deep technical credentials across the Lux investment team. | |
| New York City's Emergence as an AI Powerhouse | 4 | 7 | 2 | 2 | Molly prompts Isford on the NYC AI ecosystem. Isford presents extensive data points on Fortune 500 density, academic labs, and talent migration, while Molly humorously pushes back on moving back from the West Coast. | |
| AI Infrastructure vs Applications and Valuation Multiples | 5 | 6 | 1 | 1 | Molly asks about the shifting power dynamics between infrastructure and applications and questions whether 100x multiples have normalized. Isford distinguishes between model commoditization and durable tech differentiation, gently nuancing Molly's assumption regarding revenue drivers. | |
| Defining AI Agents and Enterprise Integration Challenges | 4 | 6 | 0 | 0 | Molly references a summit focus group on AI agents. Isford defines true agentic autonomy versus basic chatbots and explains why back-end enterprise plumbing and data orchestration remain the biggest bottleneck to scale. |
Statements from this episode (15)
Isford: 2021-2022 AI investment sectors offer little opportunity today
“The map of how this evolved, particularly in AI today, is moving so quickly that what was a very great place to maybe be investing in twenty-twenty-one, 22, is really not the place where there's a ton of opportunity today.”
Isford: Data infra teams, not AI teams, drive enterprise LLM adoption
“A lot of those data leaders are now the same people adopting AI and LLMs, and it's not like an AI team, it's just the data infrastructure tech team, right?”
Isford: AI stack challenges mirror early data infrastructure stack problems
“A lot of the same problems we saw with that early data stack, we're now seeing evolved or in parallel right with the AI stack today. So data reliability, like, is the data reliable? The same question people are asking for large language models, right? Or is it…”
Isford: Sakana AI is in many ways Japan's DeepMind
“And now Sakana in many ways is kind of like the deep mind for Japan.”
Isford: Lux invested in Hugging Face for its community and transformer unlock
“What really predicated our original investment was this incredible community. Their GitHub repo that took Advantage really of the key moment of the transformer paper unlock, productizing that, and then kind of becoming in many ways like the toll booth of AI in…”
Isford: Majority of Lux Capital's AI portfolio is based in NYC
“Majority of the Lux AI portfolio is here. I think we have a pretty darn good portfolio but that means Hugging Face is headquartered here. Runway is headquartered here. Mosaic ML, which was acquired by Databricks, They, their research team is largely here. Osmo…”
Isford: NYC captured 20% of AI VC funding, second to SF's 40%
“SF has an amazing density of talent and it's a great hub. But New York is a clear second, and it's growing in momentum. And we're seeing that again in the numbers, right? So as if I had a lion's share of AI venture capital of like 40%, I think it was over the …”
Isford: 57 of 179 relocating Lux portfolio employees moved to NYC
“We've seen like an influx of like, I think it was roughly 179 employees of our Lux portfolio companies Who moved geographies between January, 20, 20. So just before the pandemic and mid April, 20, 24. So a few months ago 57 of those one 79 moved to New York an…”
Isford: AI Models Will Become Commoditized Over Time
“Because I do think models in general will become commoditized over time, in particular because of the open source growth and adoption.”
Isford: Together AI Has the Market's Fastest Inference Tech
“They have the author of Flash Attention, Sri Dao, who is brilliant, as their chief scientist. He is, has a proprietary version of the fastest, like, inference technology on the market that's low cost to serve.”
Isford: Early-Revenue AI Startups Still Command 100x Multiples
“And so if you're an AI portfolio company you're going to get a better premium. That could change, but we're still seeing, you know, high valuations and high multiples hundred X range for companies that are earlier in revenue.”
Isford: Improved AI agent architectures will emerge within 12 to 18 months
“I think there'll be better architectures and better data collection ways to make AI agents more productive. And I think we will see that. Out of top research labs and top startups in the coming, you know, 1218 months.”
Isford: AI agents only succeed at scale when narrowly verticalized
“I think where we have seen it actually working at some sort of scale would probably be coding, customer support, or sales. And generally the ones that have worked better are very verticalized. So they're doing a clear workflow really well. They're not a genera…”
Isford: Enterprise AI failures mostly stem from implementation issues
“I think it's very easy to buy a shitty product that maybe will not solve your enterprise problem because a lot of the issues are actually on implementation of the technology.”
Isford: Enterprises should treat initial AI implementation as enterprise search
“I would view AI as an enterprise search problem. So where do you have huge repositories of data? Like what are you putting that in your data warehouse if you have one? Or where do you have the most data in your system? Or if you could search that data really q…”