🔥The Agentic Fund: How Leading… | Data Driven VC Summary

1. Key Themes

Theme: Operational AI is becoming table stakes, so the edge shifts to attention and judgment

AI adoption is already high, with a wide spread between leaders and laggards

Private market investors are past experimentation, but adoption is uneven.

"Most private market investors are well past the experimentation phase. Max puts the industry at roughly a 6 out of 10 on AI adoption, ahead of most corporate teams, with a wide spread between the leaders and everyone else."

The differentiator is no longer access to AI, it's how you direct it

"Their shared conclusion: operational AI is quickly becoming table stakes, the way spreadsheets once did. The edge now comes from where you choose to point your attention and how well you protect your own judgment."

Theme: Portability and platform independence for agent infrastructure

Store instructions, not data, and stay model-agnostic

Yohei keeps the firm's "operating system" in a form that isn't locked to any vendor, which lets him swap models freely.

"Yohei stores his firm's instructions in GitHub repos, with very little data inside. Ask any of his agents for his current LPs and it follows the trail from his chief of staff repo down to the exact CRM list ID and query. That lets him swap between Claude, ChatGPT, and Gemini at will, without locking into any single platform."

Build the portable layer yourself, buy the specialist layer

"He leaves specialist work to specialist providers like Harmonic, since replicating what they do for so many VCs would cost more and run less reliably."

Theme: Agents as always-on research and sourcing coverage

Autonomous research agents pointed at industries you're curious about

The value comes from combining novel datasets and testing the agent's predictions.

"Run autonomous research agents that go deep into the industries you care about. Combine datasets that have never been combined, have agents make predictions, then judge those predictions."

Background market maps that alert you only on change

"For ongoing coverage, map a market such as vertical legal tech and refine the list until it reflects how you see the space. Then let background agents in tools like Harmonic watch several markets at once and ping you only when something new happens."

Theme: Making proprietary data agent-ready, with provenance and security

Start with the agent, and protect your data

Firms sitting on large piles of unstructured materials can use coding agents to design the setup.

"If you have 1,000+ unstructured decks and memos sitting on a shared drive, start by asking Claude Code or Codex for setup options with pros and cons, and let it ask about your technical level first. Turn off model training on your data before connecting anything, and always ask for provenance, since a founder's own figure and a web-triangulated estimate deserve different levels of trust."

MCP connectors have a coverage gap that indexing can close

"MCP connectors handle research on one company well and broad searches across your whole network poorly. Indexing internal data against a global dataset, as Harmonic does with relationship graphs, email, and CRM data, is one way to close that gap."

2. Contrarian Perspectives

Contrarian view: Don't ask the AI whether something is a good investment. Form your own thesis first, then have models attack it. The consensus use of AI is to generate and summarize memos. The speakers argue that this erodes judgment and that polished output is a trap.

"Polished memos are now easy to generate and tempting to pass along unread. At Earlybird, we push investors to spend a few minutes on every new opportunity defining their own view and key questions upfront. Then ask several models to strawman it."

The reasoning is that leading questions produce sycophantic answers, so the framing of the prompt matters as much as the model.

"Yohei's prompt swaps 'Is this a good investment?' for 'What are the core assumptions that would have to be true to make this company successful?' Watch your phrasing, though: a leading question will almost always get you a convincing answer."

Contrarian view: Don't build everything in-house, even if you're a technical builder. Yohei created BabyAGI and runs his firm on agents, yet he deliberately outsources specialist data work.

"He leaves specialist work to specialist providers like Harmonic, since replicating what they do for so many VCs would cost more and run less reliably."

The evidence: the cost and reliability of replicating a provider's scale. Note that Harmonic sponsors this issue and its CEO is a guest, so this claim comes from interested parties.

3. Companies Identified

Untapped Capital

Description: VC firm led by Yohei Nakajima.
Why mentioned: Case study of a firm run on a portable agent operating system.
Quote: "Yohei Nakajima, General Partner at Untapped Capital, a builder by night and VC by day who created BabyAGI, one of the earliest open-source autonomous agents, and runs his firm on a portable operating system for agents."

Harmonic

Description: Startup discovery platform with an agent called Scout; sponsor of the newsletter and employer of Max Ruderman.
Why mentioned: Data backbone for sourcing, research, and background market monitoring.
Quote: "Harmonic, the startup discovery platform we use as our primary data backbone at Earlybird, which tracks the startup world in near real time and gives investors an agent, Scout, to source, research, and act on companies."

Earlybird

Description: VC firm with which the author is affiliated (referred to as "we").
Why mentioned: Case study for call preparation discipline and thesis-first screening.
Quote: "At Earlybird, we schedule 15 minutes before each call to prepare, and founders told us we came across as the best-prepared investor."

BabyAGI

Description: Early open-source autonomous agent created by Yohei Nakajima.
Why mentioned: Credential establishing Yohei's expertise.
Quote: "created BabyAGI, one of the earliest open-source autonomous agents"

Claude / ChatGPT / Gemini (Anthropic, OpenAI, Google)

Description: Frontier model providers.
Why mentioned: Swappable models in a platform-agnostic agent setup.
Quote: "That lets him swap between Claude, ChatGPT, and Gemini at will, without locking into any single platform."

Claude Code / Codex

Description: Coding agents.
Why mentioned: Recommended starting point for making internal data agent-ready.
Quote: "start by asking Claude Code or Codex for setup options with pros and cons, and let it ask about your technical level first."

GitHub

Description: Code hosting platform.
Why mentioned: Storage layer for Yohei's portable agent instructions.
Quote: "Yohei stores his firm's instructions in GitHub repos, with very little data inside."

Goodwin

Description: Law firm co-hosting the DDVC x Slush Breakfast.
Why mentioned: Event partner (Helsinki, Nov 18).
Quote: "DDVC x Slush Breakfast in Helsinki Nov 18 with Goodwin & Harmonic"

Accel, Atomico, Balderton, BlackRock, Index, Lightspeed

Description: Investment firms.
Why mentioned: Sources of recorded AI and automation deep dives in The Lab.
Quote: "100+ other AI and automation for VC deep dive recordings from leading investors at Accel, Atomico, Balderton, BlackRock, Index, Lightspeed, and many more."

4. People Identified

Yohei Nakajima

Description: General Partner at Untapped Capital; creator of BabyAGI.
Why mentioned: Practitioner running a firm on agents with a portable operating system.
Quote: "Yohei's highest-ROI agents are just as practical: one tells him what to work on each day, and another tracks every intro he has suggested and which ones actually happened."

Max Ruderman

Description: CEO of Harmonic.
Why mentioned: Provides an industry-wide view of AI adoption.
Quote: "Max puts the industry at roughly a 6 out of 10 on AI adoption, ahead of most corporate teams, with a wide spread between the leaders and everyone else."

Andre Retterath

Description: Author of Data Driven VC; investor at Earlybird.
Why mentioned: Host of the roundtable and source of Earlybird's operating practices.
Quote: "I was joined by: Yohei Nakajima... Max Ruderman..."

5. Operating Insights

Block 15 minutes of prep before every founder call and automate it

A short, repeatable prep skill is framed as a deal-winning move in competitive processes.

"A simple call prep skill surfaces which competitors raised recently and key details on the founder in about 20 seconds. At Earlybird, we schedule 15 minutes before each call to prepare, and founders told us we came across as the best-prepared investor. That moment is often where competitive deals are won."

Build small, practical agents for your own workflow first

The highest-ROI agents are mundane: daily prioritization and intro tracking, plus coding-agent benchmarking for technical diligence.

"Yohei's highest-ROI agents are just as practical: one tells him what to work on each day, and another tracks every intro he has suggested and which ones actually happened. For technical diligence, a coding agent benchmarks a startup's open-source repo against two similar projects."

Keep your playbook in portable, data-light repos

Store instructions and pointers rather than data so you can switch models and avoid vendor lock-in.

"Ask any of his agents for his current LPs and it follows the trail from his chief of staff repo down to the exact CRM list ID and query."

6. Overlooked Insights

Provenance should drive how much you trust a data point

The article only briefly notes that sourcing quality matters, but this has implications for any AI-assisted diligence: self-reported and AI-estimated figures should not be treated the same.

"always ask for provenance, since a founder's own figure and a web-triangulated estimate deserve different levels of trust."

Intro tracking as a closed-loop agent

Tracking which suggested introductions actually happened is a small detail that points to measuring the follow-through of value-add, something most funds do not systematically do.

"another tracks every intro he has suggested and which ones actually happened."