Feb 5, 2026 · 1h 10m · sourcery
How Anthropic’s $100M Anthology Fund Works | Menlo Ventures · 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 Deedy Das, Partner at Menlo Ventures, to explore modern venture capital strategies, technical AI interpretability, and founder-driven conviction. Deedy shares candid insights on why traditional investment theses fail, how macro demographic trends demand unsexy AI automation, and what it takes to build authentic influence in Silicon Valley.
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 16.4% of the talking time here. How this is scored →
Molly as informed peer 4.7 Guest teaching 5.5 Guest disagreement 2.1 Molly pushing back 1.5
speaking balance: gold is Molly, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:46 VCs Semi-Lying About Theses
DD bluntly dismisses the industry standard premise of VC theses, arguing that almost no iconic company in VC history ever originated from a thesis area.
Hardest push from Molly ▶ 40:20 Steel-Manning AI ValuationsMolly challenges DD's skepticism of high valuations by steel-manning why competitive dynamics force funds to pay whatever premium is necessary to win outlier companies.
Biggest teaching moment ▶ 48:29 The Technical Evolution of AI ModelsDD breaks down the progression of AI architectures from internet-scale pre-training and scaling laws to RL bottlenecks and test-time compute.
Molly holds their own ▶ 37:00 Carta AI Premium MetricMolly demonstrates market fluency by citing specific Carta data regarding the 30% valuation premium for AI-enabled companies and probing how to measure differentiation.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Molly as informed peer | Guest teaching | Guest disagreement | Molly pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome, Partner Promotion, and Betting on Goodfire | 4 | 5 | 1 | 1 | Molly kicks off the interview warmly, congratulating DD on his promotion to partner and asking about his biggest bet. DD explains mechanistic interpretability and Goodfire, keeping details secretive while Molly playfully prods. | |
| Macro AI Thesis: Demographics and Boring Labor Markets | 4 | 6 | 2 | 1 | DD outlines his macro thesis regarding falling birth rates, boring labor markets, and the gap between Silicon Valley hype and real-world labor shortages. Molly chimes in with humorous commentary about consumer cycles. | |
| Social Media Strategy and Building Authentic Influence | 4 | 5 | 2 | 1 | Molly asks DD how he built his massive following on X. DD explains his philosophy on writing, avoiding vanity metrics, and shares an anecdote about helping a student get into Penn. | |
| Sponsor Segment: Brex Financial Stack | 5 | 3 | 1 | 1 | Following the sponsor read, Molly and DD bond over chart-making tools and data presentation styles, comparing Claude coding skills, Freeform, and Google Slides. | |
| Anthropic Anthology Fund vs. Menlo Venture Strategy | 5 | 6 | 1 | 1 | Molly prompts DD to explain how he divides his time between the $100M Anthropic Anthology Fund and Menlo's main fund. DD clearly articulates the three-pillar strategy of the Anthology Fund and Menlo's low-volume, operator-heavy model. | |
| Demystifying VC Theses and Founder Conviction | 5 | 7 | 4 | 2 | DD challenges conventional venture wisdom, stating that VCs semi-lie about having rigid theses and emphasizing founder conviction and product taste. Molly presses him on how he tests founders before DD explains his evaluation criteria. | |
| Sponsor Segment: Turing AI Training | 4 | 3 | 2 | 1 | After the Turing ad read, Molly and DD discuss viral product growth through the story of Cluely / Interview Coder, with DD explaining why he publicly endorsed it despite controversy. | |
| AI Valuations, Hype Cycles, and Fund Discipline | 6 | 6 | 3 | 4 | Molly cites Carta valuation data and questions market dynamics around AI premiums and mega seed rounds. DD breaks down the dangers of over-raising and the 'fifty dollar lunch problem,' while Molly offers a steel-man perspective on winning top deals. | |
| Sponsor Segment: Public.com Generated Assets | 6 | 8 | 2 | 2 | Molly asks DD to map out the evolution of AI infrastructure from data labeling to reinforcement learning. DD provides an in-depth technical overview of scaling laws, RL bottlenecks, and the economic Turing test before discussing tech layoffs. | |
| Sponsor Segment: Deel Global Infrastructure | 4 | 6 | 3 | 1 | DD shares a viral insight about big tech engineering bloat and reflects on the mentorship and operating principles of Arvind Jain at Glean. Molly wraps up the episode praising the wisdom shared. |
Statements from this episode (23)
Das: Current AI explainability techniques are empirical, not fundamental
“All explainable techniques for AI today are kind of not, are, they're empirical. So they're looking at the results, they're evaluating it. That, that's not fundamentally explaining why a model does what it does.”
Das: Silicon Valley ignores critical industries to build elite white-collar AI
“The gap between like what the Valley thinks about tech and AI and focuses on and what the world cares about, it's pretty vast. And I think that cap is like not even close to being filled. So when it comes to things like insurance, we talk about what the profes…”
Das: Growing on X requires enduring public hate and cancellations
“Because you can't do Twitter without getting canceled a few times. You can't do Twitter without getting hate a few times, without people saying the meanest things about you for the world to see. You can't, you just can't do it without it. And no, most people d…”
Das: Big Tech employers repeatedly threatened to fire him over public tweets
“I couldn't do it when I was in big tech because I got a lot of pushback. They threatened to fire me a couple of times for Brand reasons.”
Das: Peter Walker joined Carta primarily for its data access
“I actually met Peter in person recently and he told me he literally just took the job at Carta because of the data access that he would get.”
Das: Uses Claude Code and custom prompt skills to generate all charts
“I am so fortunate to have like Claude code to be able to do all my charts right now. So I write myself, I've written a big skill for how I like things to be graphed, and I'm very clearly opinionated about that, like exact color and what the themes have to be a…”
Das: The Anthology Fund with Anthropic is a $100M fund
“The Anthology Fund is a fund that we do with Anthropic. It's a hundred million dollar fund.”
Das: Anthology Fund checks range from $100K to round leads
“We do usually the minimum we do is a hundred K, but we go all the way up to leading those rounds.”
Das: Doing over three deals per partner annually prevents effective portfolio support
“If you're doing five plus deals per partner a year, definitely five, but I would argue like even three or four at some .2 years in, you're going to have like six companies that you're heavily involved with, and you're not really going to have time to do work f…”
Das: Almost no iconic startup has originated from a VC's investment thesis
“Almost no iconic company in the history of venture capital has come from anyone's thesis area.”
Das: Academic institutions can no longer afford to fund frontier AI research
“Research is not happening in academic institutions anymore. They just don't have the money to fund frontier research.”
Das: Enterprise products capable of PLG must adopt it or get destroyed
“If your enterprise product can be PLG, then it must be, otherwise a PLG company will absolutely eat your lunch and destroy you.”
Das: Tech interviews are easily gamed and fail to test product skills
“I hate tech interviews with a passion. And the reason I hate them is because I've seen so many people that are just like, I call them hardcore, like they're strivers. All they do is game the interview process and they've had fantastic tech careers. And I'm lik…”
O'Shea: Carta report shows AI Series A rounds get 30% premium
“Carta put out a recent report that Series A's that are AI enabled, or AI in their name, are getting a 30% premium.”
Das: At least 90% of AI startups are just modifying LLM prompts
“And if you prod deep enough, I would say for at least 90% of companies, they're actually just, like, changing a prompt on an LLM, right? And then they're pitching all of this other stuff that they hope exists or want to believe. Could be the future of what the…”
Das: Over-raising capital erodes startup hunger and hinders recruitment
“Because if you raise too much money, we've seen companies where this happens to, A, there's an illusion of success. So the people aren't hungry anymore. We're already successful. We're a billion dollar company. Can we even fail? Right? Like that's one. Number …”
Das: Founder secondaries in over-raised venture rounds create toxic incentives
“Some people also over-raise because, you know, founders get secondary when they over-raise. So that's just the whole other topic. Generally, really poor incentives in the Valley. You know, we've seen that happen. We've heard rumors about all sorts of things wh…”
Das: Arbitrary LLM scaling fails once internet training data is exhausted
“It turns out that the scaling laws are proportional to the amount of data you can train on. So you can't just arbitrarily make the model super big if the data set is just the internet. Yeah, maybe you can accentuate that with other kinds of like proprietary in…”
Das: Reinforcement learning is an inefficient paradigm requiring massive sample sizes
“One is RL's kind of a shitty paradigm to learn. Karpathy obviously talks about this a lot. It takes a lot of samples to learn some very basic stuff because you only get a reward at the end. You don't actually understand things as it's happening.”
Das: AI models are unreliable because they lack core structural understanding
“There are many cases where models Should be more deterministic if they understood the principle involved. But they're not. They're very undeterministic because they have no core structural understanding of things.”
Das: Big Tech engineering follows a power law with long unproductive tails
“There's like a power law of who gives value in terms of engineering. There's some people who do so much that keep the company alive in any good successful big company. And there's a long tail of people who basically do nothing.”
Das: Glean received and rejected a massive early acquisition offer
“At one point we had a serious acquisition offer that we were considering. It was a big acquisition offer. We would have all made out very well.”
Das: Early startups should focus solely on whether customers love the product
“At any given point, you have one question you're trying to answer, and you want to go get that answer. That's all that you need to make something work. Don't overthink it. The one question we have right now is not any of that shit. All we need to know is do cu…”