E418: AI, Venture Capital & the Future of Investing

How I Invest with David Weisburd - David Weisburd

In this episode of 'How I Invest with David Weisburd,' the discussion centers on how artificial intelligence is transforming venture capital

Key takeaways

• AI is enabling investors to conduct deep, personalized research on founders and companies by synthesizing internal notes, public data, and industry-specific insights.
• LLMs are helping generalist VCs rapidly gain expertise in niche technical areas like semiconductors or AI chips without requiring prior domain experience.

Main topics

• AI in venture capital research and due diligence
• Use of LLMs for rapid technical onboarding

Notable quotes

"Every time I meet an investor, I'll ask them, hey, what are you using AI for? And oftentimes you get a different answer every time."

"LLMs are really good at helping people get up to speed in new technical areas... enough where they can actually follow along the narrative arc of a company that might have been just way too technical for them even a few years ago."

Conclusion

AI is reshaping venture capital by democratizing access to deep research, accelerating learning across technical domains, and enabling more

Transcript preview

Speaker 1 (0:00) Everyone says AI is transforming companies, but you think AI is going to transform venture capital itself. Why? Speaker 2 (0:07) We're seeing people experiment with new workflows across the entire lifecycle of investing from the research, sourcing, the earliest stages of identifying companies to invest in. Speaker 2 (0:26) diligence, and then all the way through to portfolio management and back office. There have been some companies that have been built to assist with various parts of that, but a lot of it right now is grassroots. And so it's just really interesting. Like every time I meet an investor, I'll ask them, hey, what are you using AI for? And oftentimes you get a different answer every time. I think it's actually a really exciting time to be an investor because people are kind of redefining what it means to be an investor in the age of AI and trying lots of new things. Speaker 1 (0:53) What's the most effective way that you know that venture firms are using AI today? A Speaker 2 (0:56) firm is going to meet with an entrepreneur. It's a perfect kind of deep research type use case of, hey, we're going to sit down with this entrepreneur, put together a full dossier on everything that this person's ever worked on, built, professional history. Speaker 2 (1:14) go do research on all of the things that have ever been published about his or her company. Another one that we see a lot is that many firms are generalist firms and or people at the firms cover many different categories. It's pretty rare these days that you'll have someone who's just focusing on semiconductors or just focusing on fintech. Of course, it does occur. But a lot of people are dabbling in a number of different areas and forced to go in many cases a mile wide and an inch deep. I think LLMs and, you know, combined with internet research, LLMs are really good at helping people get up to speed in new technical areas. So for example, if you're meeting a, you know, a new AI chip company and you don't come from a semiconductor background, the ability of one of these models to go to conduct a lot of research and tailor explanations for you for what you might be hearing from a company that meet your level of technical abilities is pretty incredible. And so we're seeing a lot of investors rely more on LLMs to help them get smart fast on new technical areas. Probably not enough for them to get all the way over the line with an investment, but enough where they can actually follow along the narrative arc of a company that might have been just way too technical for them even a few years ago. Speaker 1 (2:33) You mentioned deep research, getting ready for a meeting. What are some of the specific tools that VCs are using in order to be the most ready for those meetings? Speaker 2 (2:44) Maybe three sources of data that we see people rely on. The first is internal, let's call it proprietary data. That might be notes that have been taken, that might be stored in Notion. Speaker 2 (3:05) Previous meetings that have occurred, CRM related notes, the company being mentioned in various different memos internally. Actually getting all of that kind of all in one place and organized well is a non-trivial task. Although I think the advent of MCP has made that easier and easier for people. People connect to Affinity's MCP and they'll connect to Notion's MCP and they'll connect to Salesforce MCP or whatever tools they use. So that's one source is just all this internal information. Speaker 2 (3:34) Another source is information that's just on the internet news articles social media Speaker 2 (3:42) Wikipedia pages, whatever it might be, published academic papers. Information that's always been accessible to people, but it would have been very challenging to go and crawl through all of that information very, very quickly. SEC filings, right? There's just tons of different sources of data on the internet. And then the last one is industry-specific data products that have been built. So take like a... harmonic or a specter, maybe like two examples that are more focused on sourcing data, where they've gone and they've taken oftentimes public or public-ish data and they've built like derivative products on top of it that might show you things like, here's a list of founders that are in stealth mode, according to LinkedIn, that came from very interesting other startups. So they've kind of put a layer of almost like judgment on top of that data. And so the combination of the internal data, the external raw data, and then some of this more industry-specific curated data, I think makes for... really interesting research opportunities that let people walk into meetings with a completely different level of knowledge of what they're walking into than they did previously. It could be true for an LP meeting too, by the way. Like if you're sitting down with an LP, you know, if that LP happened to have been on a podcast or, you know, written a paper 15 years ago or whatever, you'd be much more likely to understand that and be able to speak to that. or have an interesting conversation about that going in. So I don't think it's just impacting the VC to founder relationship. I think it's also impacting the way that VCs interact with other investors and LPs and other constituents as well. Speaker 1 (5:22) And are venture firms using clot agents? Are they just putting in a prompt into ChatGPT or? We Speaker 2 (5:30) see a lot of firms that use, that are, you know, quote unquote, clod shops. We see firms that use chat GPT kind of enterprise level across the whole firm. We see firms that do both. We also see firms that use other tools like Perplexity and Gemini and the kind of Google, you know, family of products. Oftentimes it comes down to individual investor preference. And because a lot of the people that are putting together these workflows are doing so in more of an experimental fashion, it oftentimes is very path dependent on like how that person got into AI. You know, if you happen to have a friend who worked at Anthropic or something like that and got into using Claude, then maybe you would build that workflow in Claude, but someone else in the firm might build it in a completely different tool. So we're still at that place where Speaker 2 (6:19) relatively few of these workflows are enforced in a top-down manner or kind of some of them are shared and discussed of course but oftentimes it's very driven by the individual and what he or she's kind of comfortable using in the capabilities of these different tools for things like I just discussed some of maybe some of these research oriented tasks are they're all quite good at doing stuff like this so it's a little bit maybe more just that the preference of the person and then what tools the firm has approved from a compliance perspective. Speaker 1 (6:47) Full disclosure, I'm an investor in Anthropic. Congratulations. That being said, when I look at these revenue numbers, $60 billion revenue run rate, and I talk to people in the industry across different verticals, and they're all saying a similar thing, which is most of the AI usage within their organization, outside of developers, they're kind of more mature in their use, is very grassroots, individuals using their own tools. Speaker 1 (7:12) Most organizations have not actually institutionalized or operationalized this as a firm capacity. Do you see any venture firms operationalizing this? And if so, how common is that? Speaker 2 (7:25) That's a really good question. I have seen some, I think that the workflows, especially for these general purpose tools, like a cloud where you can literally ask cloud to kind of do anything you want. That's part of the beauty of it, but it's also part of the challenge. There's this sort of blank screen problem that you have where you infinite Speaker 1 (7:48) possibilities, infinite Speaker 2 (7:49) possibilities. So for some of these horizontal tools, I see a lot of the workflows that are being developed start bottoms up. And then sometimes people realize how useful they are and then they start becoming more broadly established and enforced. For example, there's one person I talked to who had a very specific process that he ran through around effectively red teaming investment memos. So like once an investment memo was published by someone in the firm, they would effectively go red team that investment memo and they would use the LLM to go in. poke as many holes as possible in the logic and the strategy behind the investment. And then the firm would kind of come together around that, and that would help provide some of the structure for their team conversation. It wasn't the only thing that they talked about, I think, but it was, hey, look, here are these four major themes that came up as we red-teamed this with the LLM, and also here are these other topics. And I think that was an example of one that was really valuable, and then it just started becoming kind of like a policy across the firm. to work that way. It became a step in the process of let's make sure that every investment memo is kind of running through a similar process. Some of the more vertical specific tools that are built for venture capital and standard metrics might be one of them. Oftentimes those are ones that are easier to go and build kind of like full institutional buy-in from day one because they probably go and fulfill a certain very specific workflow that the firm needs to do. For example, like portfolio reporting. Speaker 2 (9:19) So we're seeing a little bit of different behavior where there's some tools that are coming top down and then a lot of grassroots workflows that are being built bottoms up. And I think the interesting place is actually the intersection of those two, where because a lot of companies are now building very agent-friendly connectors like MCP, for example. Speaker 2 (9:41) A lot of firms are starting to go in, quote unquote, like vertical AI tools for their firm. But then they're also starting to go and build a lot of their own very bespoke workflows and even kind of software on top of those. So we're sort of seeing a layering now of those two. Speaker 1 (9:53) MCP model context protocol. Tell me about these MCPs and what exactly do they accomplish? Speaker 2 (9:59) The way I think about MCP is as a layer that sits on top of your application programming interface or API that makes it really easy for external large language models, AI tools, agents to go and interface with your product but allows the user to stay within the realm of prompting. So for example, Speaker 2 (10:23) We have an API for our product, and there's a way of querying the API as a customer where you could say, hey, I want you to return this data point at this time for this company. It's a function, a programming function, and there's a bunch of inputs, and there's an output that's received. It's very structured, and it needs to be done in a very specific way. What an MCP allows you to do is a layer of abstraction above that where you could go to Claude, and you could say, hey, Claude. Build me a report on this company, and it'll go in and it'll do all the appropriate underlying API calls. It'll perform actions on top of all of that data, and it'll present it to you. So it makes it much easier for a human being to use an API. Speaker 1 (11:06) It kind of answers the questions to how software firms could continue to stay relevant with this advent of these large LLM models and how they could partner with the LLM models. Speaker 2 (11:16) Sometimes the phrase that people are using these days is, you know, headless. I think Mark Benioff used that. The idea that, you know, you may not want to interface directly with the product. You may want to stay at the level of prompting or working in an AI tool. The underlying work on organizing the data, the ontology of the data, the surfacing of the data, the heavy lifting might be done by another product. But it's like a little bit of an iceberg where you're experiencing the tip of the iceberg and there's all this stuff happening below the surface. And for a lot of users, it's very empowering. For example, our product, it's a data-rich product. And so sometimes there might be a user who... needs to get something from standard metrics, but they're on the go, they're super busy, they're looking for an answer to a question. It might be a much better experience for that person to prompt an agent to grab that data for them than to log in and have a very kind of data heavy visualization type experience. There might be other users that really want to have that much more controlled experience, and so it's actually useful to be able to support both. Speaker 1 (12:27) Prior Standard Metrics, you were a VC for six years. You were at Spark Capital. If you were starting a VC firm today and you wanted to make it AI native from the get-go, what would you do? Speaker 2 (12:39) A lot of it comes down to the culture of the people because the tools are improving so quickly. Speaker 2 (12:47) And every day, week, month, year, there's so many new things that are getting launched and built that help people to be more effective at their jobs. And it's true in venture capital and private equity where we focus. But it's also true, obviously, in every other industry. And I think it's actually becoming easier and easier to adopt these tools, too. There's more. Speaker 2 (13:11) guidance, there's more services that are available. If you look at OpenAI and Anthropic, they're building out these kind of forward deployed models where they could actually go and help people to determine how to do this. There's third party service providers that will do that. Also, just people are getting better at bringing their learnings from their personal lives and other aspects of their work life and helping to kind of iterate and experiment. So I think the most important thing is the culture of the team. If you wanted to build an AI-native firm, I think you kind of need to start with people who are AI-native and or people who want to become AI-native. I've met so many people who didn't use a lot of AI tools for a long time. And then one day they decided, I want to go learn about this. And then you fast forward a couple of months and they've learned a tremendous amount. So I think the most important thing is probably the people that really desire to build that kind of firm. And then making smart decisions around process and tooling, I think, flows. from the culture of the people that are getting built. Speaker 1 (14:09) Everyone I talked to on the show is chasing the same thing, an edge. And more and more, the edge comes down to your information, not just having it, but being able to trust it when the stakes are highest. AI is doing more of the information gathering for you every day, and most tools are very good at sounding right. The summary reads clean, but can you trace it back to the filing, the transcript, the specific passage that drove the answer? Or are you just trusting the confidence of the output? For investors, that's not a minor concern. A missed filing, a misweighted source, a context that got lost somewhere in the retrieval chain, those aren't edge cases. They're how decisions go wrong. AlphaSense is the AI market intelligence platform built specifically for this. They own the content, over 500 million curated documents from broker research and expert transcripts to filings and earning calls, and they own the retrieval layer on top of it. So every answer links back to an exact verifiable source because the answer is only as good as what's underneath it. And with AlphaSense, you know exactly what that is. The edge goes to whoever could