Nov 14, 2025 · 1h 26m · latent-space
Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
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 Latent Space podcast, Menlo Ventures partner Deedy Das joins hosts Alessio Fanelli and Swix to analyze enterprise AI moats, Anthropic's rapid commercial growth, and the venture strategy behind the Anthology Fund. Das shares deep technical and strategic insights on foundation model economics, startup defensibility against frontier labs, and the future of software engineering.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 31.6% of the talking time here. How this is scored →
The hosts as informed peer 5.8 Guest teaching 6.1 Guest disagreement 2.9 The hosts pushing back 3.5
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:23:36 Cigarettes for your brain analogy
Deedy forcefully rejects Swyx's optimistic framework for fast assistive agents, arguing that developer brains will inevitably atrophy when given an addictive button to avoid thinking.
Hardest push from the hosts ▶ 1:15:39 Swyx challenges compute skepticism with exact capex figuresWhen Deedy questions whether global inference demand justifies massive capex investments, Swyx pushes back with concrete data showing OpenAI spends $2B on inference versus $5B on R&D.
Biggest teaching moment ▶ 11:31 Deedy breaks down consumer versus enterprise search physicsDeedy explains why consumer search ranking techniques fail in the enterprise due to low feedback volume, distributed queries, and freshness demands.
The host holds their own ▶ 1:22:22 Swyx outlines semi-async value death and fast agentsSwyx demonstrates deep domain expertise by articulating his technical thesis on fast versus async coding agents based on models he personally shipped that day.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome Back and the Evolution of Claude | 6 | 5 | 3 | 4 | Swyx and Alessio bring up Anthropic's enterprise search release and SaaS API rate limits. Deedy breaks down first-principles business logic for why SaaS providers shouldn't restrict APIs and why foundation labs won't dedicate engineering to commoditized connectors. | |
| Overcoming Enterprise Search and Distribution Hurdles | 5 | 8 | 2 | 2 | Deedy educates the hosts on the fundamental mechanics that differentiate enterprise search from consumer search, including the lack of query feedback volume, lack of head-heavy distributions, and non-viral adoption dynamics. | |
| Anthropic's Unprecedented Growth and Distinct Culture | 6 | 6 | 2 | 3 | The hosts and guest examine Anthropic's rapid growth and unique research-driven culture. Swyx and Alessio cite industry retention statistics and API spend surveys, while Deedy clarifies market share definitions and trajectory. | |
| Frontier Model Underwriting and Enterprise Coding Value | 6 | 7 | 3 | 3 | Deedy explains why market share is a vanity metric when underwriting frontier labs, contrasting the saturated consumer intelligence ceiling with the uncapped frontier in software engineering. | |
| Model Layer Defensibility vs the Application Layer | 6 | 6 | 5 | 5 | Deedy challenges the thesis that application layers capture all AI value, arguing model layers hold the true moat because they are harder to build. He also pushes back on the claim that Claude Code is universally regarded as superior to Cursor or Devin. | |
| Rahul Patel's Ascent and Overcoming Meritocratic Barriers | 4 | 7 | 1 | 2 | Deedy provides cultural and systemic context on Indian engineering education and meritocracy, discussing how Anthropic CTO Rahul Patel bypassed conventional elite credential pathways. | |
| Structuring the $100M Anthology Fund with Anthropic | 5 | 6 | 2 | 2 | Deedy details the architecture of the Anthology Fund, detailing why hosting it externally via Menlo Ventures avoids the standard misaligned incentives of traditional corporate venture capital. | |
| Deep Research Bets: Goodfire and Prime Intellect | 6 | 7 | 3 | 4 | Swyx probes research investments like Goodfire and Prime Intellect. Deedy explains mechanistic interpretability as brain surgery for model weights and outlines why distributed compute approaches hold upside. | |
| OpenRouter: Building the Developer-First Model Gateway | 6 | 6 | 3 | 4 | Alessio questions OpenRouter's defensibility against integrated gateway offerings like Vercel AI Gateway. Deedy counters with OpenRouter's developer-first depth, zero-retention flags, and granular provider-level metrics. | |
| Portfolio Spotlights: Wispr Voice and Diffusion Code Generation | 6 | 6 | 2 | 3 | Deedy presents the case for Wispr voice dictation and explores diffusion models for code generation, arguing bidirectional dependency resolution fits programming better than left-to-right autoregression. | |
| Market Timing, Multiple Arbitrage, and Massive AI Infrastructure | 6 | 5 | 2 | 3 | The conversation covers market timing, PE multiple arbitrage, and massive infrastructure buildouts. Swyx and Alessio discuss Databricks' Mosaic acquisition and the reflexivity of mega-rounds. | |
| The Compute Question: R&D Scaling vs Inference Reality | 7 | 5 | 4 | 6 | Deedy questions whether massive compute spending will yield proportionate economic value or inference demand. Swyx counters directly with concrete breakdown figures on OpenAI's $2B inference versus $5B R&D capex. | |
| Vibe Coding Risks, Brain Atrophy, and the Future of Engineering | 7 | 5 | 6 | 5 | Deedy warns that vibe coding degrades developer cognitive muscles, likening it to an addictive slot machine. Swyx details his newly shipped model paradigm of fast assistive agents, but Deedy forcefully pushes back with a smoking analogy. |
Statements from this episode (24)
Sam Altman Barred Investors Who Backed Glean From Investing in OpenAI
“Sam Altman once came out and said, if you're an investor in OpenAI and one of these five companies, including Glean, we don't want you as an investor.”
Glean Operates at a Several Hundred Million Dollar Revenue Scale
“Look at the revenue of Anthropic and OpenAI right now. These are billion dollar revenue scale businesses. Glean is several hundred million dollar revenue scale business.”
AI Labs Will Not Dedicate Engineering Talent to Deep Enterprise Search
“If you really want to go deep, I don't think you will ever dedicate the people to do it. And the last thing I'll say is you think about from an anthropic engineer's perspective, you joined a big AI lab to work on models, not to build Google drive connectors, r…”
Enterprises Buy Productivity Software for User Love, Not ROI
“Selling productivity tools to enterprises are challenging because as no matter what ROI argument you make, people aren't actually buying tools for ROI. People buy productivity tools because their users like using them.”
Anthropic Is the Fastest-Growing Software Company in History
“Anthropic is the fastest growing software company of all time. I think I can say that fairly. I'm, I haven't been disproven yet.”
Anthropic Maintains an 80% One-Year Employee Retention Rate
“I'm referring to the exact same article where I think their retention, one year retention on employees is the 80%, which in AI world is, is quite wild.”
Enterprise LLM Churn Is Low Due to Long-Term Compute Commitments
“In terms of enterprises, often what will happen is they'll buy up large chunks of long-term compute and dedicated instances, in which case you just don't churn, right? Like this is what you use.”
Menlo Underwrites Late-Stage AI on Revenue and Margin, Not Market Share
“At that stage, to be very honest with you, at the stage that we invest in Anthropic now, like the only things that would really move the needle on the decision is here's the revenue, here's the margin, and here's the trajectory, and here's the other markets we…”
Smarter Consumer AI Models Do Not Increase User Growth or Retention
“Furthering the intelligence of models and chat GPT, a consumer product does not lead to more users or more retention. It only is really applicable to us the thin slice of users who care about very smart type queries, right?”
OpenAI Has Effectively Won the Consumer AI Market
“Now that means we're at a point in consumer where, maybe this is too early to say, but OpenAI has kind of won, right? Like, How do you catch up to something where model quality is not going to be differentiated? You already have the users, you already have the…”
AI Coding Intelligence Has an Uncapped Frontier That Drives Revenue
“But the interesting about Anthropic is if you look at coding, that's probably never going to be the case. Like there's always an increasing frontier of how you good you could be at a task like that. And we're nowhere close to that frontier. So it's more possib…”
Model Layer Is Structurally More Defensible Than the AI App Layer
“It is far easier for Anthropic to try to go into one of the spaces of the apps than an app to try to go into the space of Anthropic, which makes me feel like one is more defensible than the other all else equal.”
Corporate Venture Funds Perform Poorly by Prioritizing Product Usage Over Quality
“Typically, if you look at corporate venture funds in history, obviously, besides OpenAI as a notable exception, they tend to not be very good because all they prioritize is who uses my stuff the most.”
Interpretability Research Will Explain AI Model Outputs Within Three Years
“I think that if we further that research direction two, three years in the future, we will be able to understand why models say what they'd say.”
Mechanistic Interpretability Scales Without Bottlenecks to Large Models Like DeepSeek
“There's no gap for scale. Like, they've shown that even for the biggest open source models, you like, even like DeepSeq's big models, you, they can do it. And then in general, like, scaling is not the bottleneck.”
In Any SaaS Market With PLG Potential, PLG Will Win
“I think go so far as to say for in any SaaS market, if there can be a PLG motion, the PLG motion will win.”
Voice Dictation Startup Wispr Achieves a Zero-Edit Rate Above 80%
“And their zero edit rate, I think is north of 80%, which is insane for a voice dictation product.”
Diffusion Models Achieve 80% of Autoregressive Quality at One-Tenth the Cost
“Diffusion models today are, I would say, 80 to 90% of the quality at one-tenth the cost and latency.”
Left-To-Right Autoregressive Reasoning Is Suboptimal for AI Code Generation
“Left to right reasoning for code doesn't actually really make sense, because in code, we don't, like, we might sometimes write code left to right, but after you write code, you go up and down and figure out, hey, is this variable set? Did I do this? There are …”
Anthropic's MCP Has Won as the De Facto AI Interop Standard
“Now it's like basically kind of de facto one as the interop layer for all the labs and all the models.”
Venture Mega-Rounds Cancel Competition and Create Market Winners by Deterrence
“There's times where people funnel so much money into a company before it's really like prime time, which dissuades anybody else from entering that market. And then they become the de facto owner of the market because they canceled the competition with funding.”
OpenAI Plans to Scale Compute Power Capacity to 125 Gigawatts
“For OpenAI to go from like two gigawatts of compute this year to 30 with everything they've already announced, and then there's a plan for the next 125. Like, the United States uses 300.”
OpenAI Spent $7 Billion on Compute, With $5 Billion for R&D
“This year, OpenAI spent seven billion dollars on compute. Only two of that was for all of their inference. The remaining five was R&D. So all of ChatGPT, all eight hundred million users, all of Sora, all of like, all, all the sort of like API volume, two billi…”