Things I've been thinking about -

I can’t believe it's the end of March. A lot has been brewing, so I thought I’d write about some things I’ve been thinking about —

Crazy Stat: YCombinator

25% of W25’s YC startup batch had 95% of their codebases generated by AI (source)

I can’t shake this stat.

We all know code generation capabilities have reached insane levels, driven by Claude 3.7 and the fact that some of the most brilliant people in the world are working in the space. So we’re spending a lot of time thinking about: what can’t be prompted? It feels like now more than ever, there is a greater emphasis on unique expertise, execution pace, technical moats, product taste, and distribution advantages for venture-backable companies.

Haunting Me: Agents

Today, I had the privilege of participating in Plug and Play’s venture panel on AI innovation.

Matt Hennigan asked as host: Is there an AI advancement on the horizon that excites you but also keeps you up at night?

The answer for me is simple: agents.

A longer response —

Early large language model applications were mostly focused on information retrieval and content generation (chatbots, basic text summarization). The tools could summarize or describe, but rarely would complete complex tasks end-to-end, even with sophisticated prompting and automation.

Aside from foundational models, code generation tools and voice infrastructure made the most money in this first era.

Now, we're in a second wave where reasoning agents are everywhere. They're top of mind for everyone — in pitches, on Twitter, within product roadmaps — we're clearly moving from one-off answers to continuous chains of reasoning and action. The goal is executing intelligent, multistep workflows autonomously (and often asynchronously in the background).

Who will make money in the agent era? Obviously, companies that deliver measurable economic value: reducing costs, speeding up workflows, and enabling capabilities that translate directly to revenue.

But from an investment perspective, what are the true paradigm shifts? I believe many agent applications within vertical AI and enterprise software will get absorbed by existing solutions as incumbents build out and integrate these capabilities. The more interesting opportunities lie in agent infrastructure and orchestration layers — but what keeps me up at night is figuring out what companies should build versus buy from the agent stack — and identifying where a new AI company hasn't already been funded but can still break through.

OpenAI’s Deep Research

I've been using OpenAI's Deep Research for investment diligence and wow wow wow.

It’s beautiful.

Early-stage moves quickly. We’re turning around most investment decisions within 48 hours. It feels truly crazy to be able to ask a specific question and receive a full, well-researched report in just 20 minutes.

An example — last night, I was looking at a company in a hot category but in an area with many graveyard companies. I asked Deep Research to analyze every company that had shut down in this space and provide an in-depth analysis of why. Being able to accelerate the pattern matching is huge - and in this case, actually made us more excited about the “why now.”

I’ve finally come to accept that early-stage investing will always be more of an art than a science, but it feels like, maybe for the first time ever, we can get surprisingly close to perfect information to make fast decisions.

Carta VC Performance Report

Carta’s VC Performance Report was just released, and there is a lot of good content there. Added a few metrics below, but full report here (link)

• In the upper tiers of performance, the smallest venture funds often post the best returns. For 2018 vintage funds, the 90th percentile TVPI for funds with just $1M - $10M in assets hit 4.03x, while funds over $100M only reached 1.67x at the same percentile.

• Half of all funds from the 2018 vintage had still not distributed any capital back to their LPs.

• Interestingly, the 2023 vintage is showing some early signs of promise, with the 90th percentile for IRR after one year being 24.5%, which is higher than the 2022 vintage and similar to figures from the late 2010s.

I deeply believe in our fund strategy, and it’s been working. Without question, we are a top percentile fund for our 2021 vintage fund. Even so, the numbers in the report are humbling.

Prompt Generation & Claude

I’ve been using Anthropic’s prompt generator to create/recreate the prompts for all of our internal data systems: https://console.anthropic.com/dashboard

It's a product I didn't know existed until Rebecca Harbeck ran a workshop during our AI Engine Hackathon.

And I've actually switched almost everything over to Anthropic models, with OpenAI's Deep Research being the only exception. Claude just seems significantly better for most things.

Product Market Fit London 🇬🇧

We hosted our first Product-Market Fit event in LA last month! It was a massive success. We had amazing conversations with industry leaders including: Nikita Bier (Partner at Lightspeed), Jack Brody (CPO of Suno), Gaurav Misra (CEO of Captions), Danny Trinh (Director of Product Design at Meta), Todd Jackson (Partner at First Round Capital), Noah Fradin (Stealth), and Daniel Edrisian (CEO of Alex)

I'm so excited to announce Chapter One's second Product Market Fit event coming to London on April 8th, 2025!! 🙌 🔥 🇬🇧

✅ 🔥 If you are in London, please join us!! https://lu.ma/qy3m1d41

We feel incredibly lucky to have an insane lineup of speakers: Harry Stebbings (Founder of 20VC), Sam Stephenson (Co-Founder of Granola), Simon Taylor (Head of Strategy & Content at Sardine), and a final speaker that we’ll announce later in the week!

Ghibli Style Anime

Ghibli-style photos are blowing up on Twitter & I couldn’t resist.

Family photo from my niece’s first birthday last weekend ❤️

I’m an investor at Chapter One, an early-stage venture fund that invests $500K - $2M checks into pre-seed and seed-stage startups.

If you’re a founder building a company, please feel free to reach out on Twitter (@seidtweets) or Linkedin (https://www.linkedin.com/in/jamesin-seidel-5325b147/).