Apr 17, 2026 · 34m · american-optimist
How AI Agents are Changing Investing · Joe Lonsdale
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 American Optimist, host Joe Lonsdale interviews John Milles-Kiriazzi, CEO of Standard Metrics, on how artificial intelligence and autonomous agents are reshaping venture capital operations, due diligence, and portfolio management. The conversation highlights John's career trajectory from Stanford physics research to founding a leading financial data platform, while analyzing the future of software moats in an AI-driven economy.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Joe holds 27.9% of the talking time here. How this is scored →
Joe as informed peer 4.9 Guest teaching 2.9 Guest disagreement 0.6 Joe pushing back 1.1
speaking balance: gold is Joe, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 24:47 Reframing software defensibility moats
John challenges the conventional SaaS moat thesis, pointing out that AI directly destroys traditional switching cost advantages.
Hardest push from Joe ▶ 27:56 Joe dismisses DCF valuation utility in VCJoe bluntly rejects John's example of Claude building a discounted cash flow, pointing out DCF is largely irrelevant in modern venture pricing.
Biggest teaching moment ▶ 29:02 Explaining the Model Context ProtocolJohn explains the mechanics and architectural power of Model Context Protocol (MCP) to automate complex cross-tool investor workflows.
Joe holds their own ▶ 29:47 Joe breaks down rigid APIs vs MCPJoe articulates his technical understanding of traditional rigid APIs compared to conversational, flexible LLM context protocols.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Joe as informed peer | Guest teaching | Guest disagreement | Joe pushing back | Why |
|---|---|---|---|---|---|---|
| Early Background, Physics, and Engineering | 3 | 4 | 0 | 0 | Joe asks friendly biographical questions about John's upbringing, basement tinkering, and PhD research in photovoltaics. John details his device physics papers on loss mechanisms in solar cells without any friction. | |
| Evolution of Academia vs. Startups | 6 | 2 | 1 | 1 | Joe asserts his philosophical views comparing academia to startups using Aristotle and Kant, and shares his own prompting hacks for literature reviews. John builds collaboratively on how AI acts as a technical co-pilot for diligence. | |
| Founding and Mission of Standard Metrics | 5 | 3 | 0 | 1 | Joe notes his firm 8VC's involvement in co-founding Standard Metrics and prompts John to explain the venture data problem. John walks through the manual data collection pain points and cross-portfolio benchmarking. | |
| AI's Impact on Venture Capital Operations | 4 | 3 | 1 | 2 | Joe questions why AI cannot simply use a common-sense check rather than requiring human QA. John clarifies the difference between structured database queries and unconstrained document extraction. | |
| Standard Metrics AI Analyst & Unstructured Data | 5 | 3 | 0 | 1 | Joe connects John's workflow to his experiences with Addepar and Vercel, while John explains internal vibe coding and reverse-engineering LLM discovery. The rapport is entirely aligned and collaborative. | |
| The SaaSpocalypse and Network Effect Moats | 5 | 2 | 1 | 1 | Joe brings up the 'SaaSpocalypse' narrative and John dissects how switching costs diminish while data and network effects remain strong defensible moats. Joe affirms how Standard Metrics' multi-firm network powers onboarding. | |
| Automated Valuations and Agentic Integrations | 6 | 3 | 1 | 2 | Joe jokes that discounted cash flow models have nothing to do with VC valuations and breaks down the difference between rigid APIs and MCP. John demonstrates how MCP orchestrates multi-app tasks like automated board prep. |
Statements from this episode (16)
Lonsdale: Startups Offer Greater Intellectual Satisfaction Than Academia
“You know, I think this is a really common thing in our society right now, where I think it used to be maybe 50 years ago, a hundred years ago, the smartest people who wanted to pursue the most intellectually demanding and interesting things would stay in acade…”
Melas-Kyriazi: AI Enables Investors to Conduct Technical Diligence Within Hours
“I think like AI as a technical co-pilot for diligence is actually one of the most interesting use cases for AI. As an investor right now. I don't think it's a replacement for human experts necessarily, but it certainly is a huge help. If you're, if you come ac…”
Lonsdale: Prompting AI to Explain Academic Papers Enables Fast Learning
“I actually, one of the hacks That I really enjoy when I have free time is to ask the AI, like, what are the most interesting and important or cited papers or new papers or controversial papers in a field? And then explain these to me as if I only have an under…”
Melas-Kyriazi: VCs Use Claude and ChatGPT to Critique Investment Memos
“Another thing we're seeing is that firms are writing investment memos, and then they're using tools like Claude and ChatGPT to poke holes in them. What am I missing? Where, where's my argument weakest? And then using that to help to guide the diligence process…”
Melas-Kyriazi: Singapore VC Uses AI Agents to Automate LP Reporting
“For example, we have one customer in Singapore that's using an agent orchestration tool, and it's building this wide variety of AI agents that interface with standard metrics, grab data, and fulfill a bunch of really critical internal reporting workflows. For …”
Melas-Kyriazi: AI Data Extraction From Unknown Documents Requires Human QA
“What's tougher is feeding in a completely unknown document and having AI extracted well out of the unknown documents. We do that currently, but we use humans in the loop to make sure that we QA. It is not a fully solved problem yet. We think it will be in the …”
Melas-Kyriazi: AI Agents Do Vast Majority of Standard Metrics Data Operations
“And the first place we launched AI agents in our product was to assist the data operations team. Now AI is doing the vast majority of the work.”
Melas-Kyriazi: Standard Metrics Tracks LLM Search Visibility With Daily Prompts
“So yeah, so we actually use software that helps us. It runs hundreds of prompts per day across four or five kind of major LLMs. It helps us to identify exactly where standard metrics is getting mentioned. Which pieces of content that are linked to standard met…”
Melas-Kyriazi: Standard Metrics Creates Marketing Content Engineered for LLMs
“We're creating content specifically for LLMs that's also human readable and useful for humans, but it's really engineered to make sure that LLMs understand what we do.”
Melas-Kyriazi: AI Has Eradicated Traditional SaaS Switching Cost Moats
“Because of AI, that's changing. It's cheaper for competitors to build software. It's cheaper for people to vibe code their own software. And it's also easier to migrate data because of AI agents. So I think switching costs have gone down. And so I think it's r…”
Melas-Kyriazi: You Cannot Vibe Code a Network or Proprietary Dataset
“You can't like vibe code a network. You can't vibe code a new data set.”
Melas-Kyriazi: Major Opportunity Ahead for AI-Automated Venture Valuations
“I think that there's going to be a lot of opportunity for innovation in, in automation around valuations, leveraging AI plus benchmarks, plus proprietary company data, as well as Public market comps and other.”
Melas-Kyriazi: Claude Automatically Generates DCF Models in Excel
“Like one thing that we published recently was you can go into Excel, hook it up with Claude, hook up Claude with standard metrics, and then ask it to do a discounted cashflow for a portfolio company. It'll just build it for you.”
Melas-Kyriazi: Claude Autonomous Task Duration Has Reached 20 Hours
“What there's a task that has a 50% chance of successful completion. How long does that task look like for a given model? I think Claude is now at like 20 hours or something like that. And it was at like, you know, one hour a year ago.”
Melas-Kyriazi: SaaS Apps Must Go Headless to Integrate User AI Agents
“Some people are referring to this as like a headless model, and I think it's really important for companies to be thinking and building with that in mind and making sure that if their users are extremely sophisticated with how they're thinking about building t…”
Melas-Kyriazi: AI Back-Office Automation Will Drive Explosion of Small Businesses
“I think what we're seeing now, which is really exciting is this explosion of small businesses getting started. And I think that people are going to be able to spend more time on what they're truly passionate about, probably spend more time on things that direc…”