Who Is Yohei Nakajima and How the VC Behind BabyAGI Spawned a Generation of AI Agents

Yohei Nakajima is the venture capitalist who, in late March 2023, posted under 200 lines of Python to GitHub under the name BabyAGI and accidentally named an entire category of software. AutoGPT, AgentGPT, Microsoft AutoGen, and the dozens of agent frameworks that followed all trace their lineage back to that one repository. Nakajima himself is a General Partner at Untapped Capital, a Venture Partner at Scrum Ventures, and one of the very few investors who is fluent in both autonomous-agent research and on-chain crypto.

For traders the name showed up again in December 2024, when Nakajima released the Pippin Framework, an evolution of BabyAGI with persistent memory and character definition. A Solana memecoin built on top of that framework, also called PIPPIN, deployed the same month and pushed to a roughly $800 million market cap inside weeks.

The VC Who Codes On The Side

Most venture investors who talk about AI agents have never written one. Nakajima is the exception. He runs Untapped Capital, an early-stage fund he founded that backs what he calls "untapped" founders, meaning underrepresented or under-networked teams building in AI, automation, and infrastructure. He is also a Venture Partner at Scrum Ventures, where he sits on cross-border deals between US and Japanese tech.

Before all of that he was an operator. He has talked publicly about working as an in-house web engineer and product manager for years before ever picking up a check book, which is unusual for a Silicon Valley GP and a big part of why his code projects read like a working engineer's notebook rather than a thought-leader's slide deck.

He is also one of the most-followed AI-and-crypto voices on X, with reporting from late 2024 and 2025 noting that his account is followed by Jeff Bezos and Marc Andreessen. That is the rare crossover audience of frontier-tech founders and crypto-native traders that almost no other investor has at the same time.

BabyAGI And The Repository That Started The Agent Wave

In late March 2023, four months after ChatGPT launched, Nakajima pushed a Python script to a public GitHub repo. It was tiny, around 140 lines in its first release. The script wired a language model to a vector database and a simple loop that took an objective, broke it into tasks, executed one, generated new tasks from the result, and looped.

He called it BabyAGI, and the repo went viral inside a week of going public.

The reason it mattered was not the code, which was deliberately minimal. It was the proof. BabyAGI showed in public, with copy-paste-ready Python, that a model could plan and act in a loop without a human in the seat. Within a month AutoGPT and AgentGPT had launched as scaled-up cousins, Microsoft research papers cited BabyAGI directly, and the term "AI agent" entered the mainstream tech vocabulary. Nakajima kept the original repo as a teaching artifact and shipped a series of follow-ups (BabyBeeAGI, BabyCatAGI, BabyDeerAGI, BabyElfAGI), each one adding a single new capability and nothing else.

That cadence is the signature. He treats open-source agent code the way a mathematician treats lemmas, one small public step at a time, each fully understandable.

The Pippin Framework And Why It Matters For Crypto

Two years after BabyAGI, in December 2024, Nakajima released the Pippin Framework. It is the next generation of the same idea, but with three additions that BabyAGI never had. The first is a character definition, meaning the agent has a persistent personality, voice, and worldview encoded as data rather than re-prompted every loop. The second is a real memory system, with vector storage that persists across runs so the agent learns from its own history. The third is external tool integration, meaning the agent can post to X, mint NFTs, send transactions, or call any API the operator wires up.

In other words, Pippin is the first easy-to-fork open-source framework for autonomous AI characters that can hold a personality and act on-chain. That last detail is why crypto picked it up so fast.

Within days of the framework release, an anonymous team launched a token on Solana also named PIPPIN, marketed as the on-chain expression of the framework's reference character, a digital unicorn. The token captured the AI-agent-coin narrative that was already running on Solana behind names like GOAT and zerebro and pushed to roughly $800 million in market capitalization at its peak in January 2025. Nakajima has been careful in public to clarify that he did not launch and does not control the token, but the framework is his and the token is built around the character he designed.

The Untapped Capital Thesis

Untapped is small by Sand Hill Road standards, with roughly $16 million across its first two funds, but the portfolio is the part that matters. The fund concentrates on pre-seed and seed checks into AI-agent infrastructure, automation tooling, and crypto-adjacent middleware, with a heavy bias toward founders who are not pattern-matched to standard YC profiles.

The thesis Nakajima has articulated in interviews and on his blog is straightforward. Autonomous software, in his view, is going to need its own infrastructure stack for identity, payment, memory, and coordination. A human signs into Stripe with a credit card and a phone number. An agent cannot do either of those things at scale. The plumbing that lets an agent prove who it is, pay for compute, store its own memory, and find other agents to work with is mostly missing, and that gap is where his check book points.

This is also the bridge to crypto, because public-key cryptography already solves agent identity, stablecoins already solve agent payment, on-chain storage and IPFS already solve agent memory, and token-gated coordination markets already solve agent discovery. None of that needs to be crypto in theory, but in practice every component already exists on-chain and almost none of it exists off-chain in a permissionless form.

What Pippin Proved About AI Tokens

PIPPIN's run to roughly $800 million in market cap is small compared to the largest memecoins, but it proved a specific point that the rest of the market is still digesting. An AI character with a persistent personality and a public open-source framework behind it can become a billion-dollar asset class in weeks, with most of the value flowing to the framework's narrative rather than to its creator's wallet.

The follow-on effect was visible all through early 2025. New AI-agent tokens launched on Solana every week, most copying the Pippin template of a named character with a personality, a Twitter account run by the agent itself, and a tradable token tied to the character. A handful, including Truth Terminal-adjacent names, crossed comparable market caps before the broader memecoin cycle cooled.

For traders the takeaway is not that the next PIPPIN is around the corner. It is that the AI-agent narrative on Solana is now a recurring category, and that the framework releases out of Nakajima's GitHub are the most reliable leading indicator of what the next cycle of agent tokens will be built on top of. When he ships, the on-chain teams notice within days.

How To Track What He Does Next

The honest answer for most VCs is that you cannot. They speak at conferences twice a year and the rest is private. Nakajima is the opposite. His GitHub is public, his X account is active and engineering-focused, and his blog at yoheinakajima dot com publishes essays on agent design with the actual diagrams. Untapped Capital's portfolio is partially public on its site, which is rare for a pre-seed fund.

The simplest tracking rule is to watch the GitHub releases. Every public repo he ships is a hint at what infrastructure the agent-token narrative will lean on six to twelve months later. The Pippin release in December 2024 was visible at least three weeks before the SOL-based AI character coins broke out. The next time a similar repository ships, the same template applies.

Frequently Asked Questions

Did Yohei Nakajima invent AI agents?

No, the research on autonomous agents goes back decades and predates BabyAGI by a long way. What he did in March 2023 was ship the first widely usable open-source proof that a large language model could plan and act in a loop with under 200 lines of code. The category existed in research, and he made it a thing tens of thousands of developers could fork and modify in an afternoon.

Is PIPPIN officially run by Yohei Nakajima?

No, and this is the detail that often gets blurred in social media coverage. Nakajima built and released the open-source Pippin Framework, which defines the character. An anonymous team launched a Solana token called PIPPIN built around that character, and he has publicly stated he does not control or run the token itself. The framework is his, the token is community-launched.

Why does an AI venture capitalist matter for crypto traders?

Because the AI-agent narrative is now one of the most durable token categories on Solana, and the frameworks driving it come almost exclusively from a small group of independent builders, with Nakajima at the top of that list. Watching what he ships open-source is a leading indicator for which on-chain agent tokens will rotate into next.

What is Untapped Capital's connection to AI agents on-chain?

Untapped writes pre-seed and seed checks into agent infrastructure, automation tooling, and AI-meets-crypto middleware. The thesis is that autonomous software needs its own payment, identity, memory, and coordination layers, and that those layers in practice end up being on-chain because public-key cryptography and stablecoins already solve problems that off-chain systems do not.

Bottom Line

Nakajima is one of the very few people who can credibly claim to have helped name a software category, and the category he named is now one of the most-watched themes in both AI and crypto. BabyAGI gave the agent narrative its template in 2023. The Pippin Framework gave it its on-chain expression in late 2024. PIPPIN gave the market its first proof that an open-source AI character can become a multi-hundred-million-dollar token in weeks.

For traders the practical signal is the GitHub. Every public release Nakajima ships sits roughly three to six weeks ahead of the on-chain teams that build tokens around the same primitives, and the pattern has already repeated cleanly once. The next time the same setup appears, a brand-new open-source agent framework with a named character, persistent memory, and on-chain hooks, the trade is to be early on the framework's terminology before the Solana launchpads pick it up. Watch the repo, not the timeline.

This article is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency trading involves substantial risk. Always conduct your own research before making trading decisions.