Yohei Nakajima is the creator of BabyAGI, one of the first popular open-source autonomous agents, and the founder of the Pippin AI project. In his venture capital career, he is a General Partner at Untapped Capital and a Venture Partner at Scrum Ventures. ^{[1] [2]}​

Education

Nakajima attended the University of Washington, where he earned a Bachelor of Arts in Business Administration. ^{[3] [4]} Other sources have also cited his studies in fields including Marketing, Finance, Information Systems, Japanese, and International Studies. ^{[5] [6]} Some professional profiles list his attendance at Claremont McKenna College, where he reportedly studied Economics, Psychology, and Leadership. ^{[2] [7]}

Career

Nakajima began his career in the Los Angeles startup ecosystem, organizing educational events and courses for founders. [[scrum.vc/team/yohei-nakajima/][Scrum Ventures team profile for Yohei Nakajima]] He was an early entrepreneur, founding multiple startups, including Mindle Inc. while in college, and later co-founding dotstud.io, a platform for filmmakers, and Moveloot, a Y Combinator-backed marketplace for secondhand furniture, which served as a basis for working with Web3. ^{[4] [8]}​

He transitioned to venture capital in 2014 by joining the global accelerator Techstars. ^[1] During his time there, he served as a Program Manager and later became the director of the Disney Accelerator, a program powered by Techstars. ^[7] He also held the position of Director of Pipeline, where he managed worldwide startup sourcing and recruitment for over 30 accelerator programs. ^{[2] [9]}​

In 2018, Nakajima joined Scrum Ventures as a Venture Partner on the investment team. He also became the SVP of Scrum Studio, the firm's corporate innovation division, where he designed and led programs with global corporations, including managing the partnership with Nintendo. ^[1] In December 2020, he co-founded Untapped Capital, an early-stage venture capital firm, with Jessica Jackley. The firm utilizes AI and automation in its operations and follows the "build-in-public" approach that Nakajima champions. ^{[1] [10]}​

In addition to his investment roles, Nakajima serves on the boards of E. A. Renfroe & Company, an insurance claims management company, and Misapplied Sciences, a company developing advanced display technology. ^{[7] [2]}​

Philosophy and Methodology

Nakajima's work is guided by two central philosophies: "VC by day, builder by night" and "build-in-public." The former encapsulates his dual role as a hands-on software developer who experiments with new technologies, primarily at night, while working as a venture capitalist during the day. ^[1]​

His "build-in-public" methodology involves openly sharing his experiments, progress, and code through his personal blog and social media. He maintains a public "build log" to document his projects in AI, no-code, and Web3. ^[11] Nakajima has stated that this transparent approach helps him "learn faster, meet founders, and contribute to the ecosystem." ^[1]​

Major Works and Projects

BabyAGI

BabyAGI is an open-source, task-driven autonomous AI agent created by Nakajima in March 2023. It became one of the first and most influential projects in the autonomous agent space, demonstrating how a Large Language Model (LLM) could be used to autonomously manage and execute complex tasks. ^{[2] [3]}​

Origin and Concept

The project originated as a thought experiment to create a simplified, task-driven autonomous agent. The name "BabyAGI" was chosen to signify a small, initial step toward Artificial General Intelligence, not to claim it was AGI. ^[2] The initial Python script was approximately 140 lines of code and was published alongside a blog post titled "Task-driven Autonomous Agent utilizing GPT-4, Pinecone, and LangChain for diverse applications." ^[12]​

The system operates in a continuous loop based on a user-defined objective:

1. It pulls the highest-priority task from a task list.
2. It sends the task to an execution agent, powered by an LLM like GPT-4, which completes it based on the objective and context from previous results.
3. The result is stored and enriched in a vector database, such as Pinecone, which serves as the agent's memory.
4. A creation agent generates new tasks based on the objective and the recent result, and a prioritization agent re-ranks the entire task list. ^{[8] [12]}

Impact and Evolution

BabyAGI went viral on GitHub and Twitter, accumulating over 18,000 stars and millions of impressions shortly after its release. Its simplicity and open-source nature inspired a wave of similar projects, including Auto-GPT, and it has been cited in numerous academic papers. ^[1] Following its initial success, Nakajima released several experimental variants, including BabyBeeAGI, BabyCatAGI, and BabyFoxAGI. ^[13] In September 2024, the original project was archived and relaunched with a new focus on being a "self-building autonomous agent"—a minimal framework designed to build and improve itself. ^[3]​

On the project's repository, Nakajima includes a disclaimer that highlights his product-first approach: "This is a framework built by Yohei who has never held a job as a developer. The purpose of this repo is to share ideas and spark discussion and for experienced devs to play with." ^[3]​

Pippin

Pippin is a multifaceted project that evolved from Nakajima's work with AI agents, combining elements of an AI character, a Web3 social game, and a modular AI framework. ^{[2] [4]}​

Origin

The project began serendipitously in October 2024 when Nakajima used an AI model to generate an SVG image of a unicorn, which he named "Pippin." After he shared the image on social media, a community member launched a meme coin named PIPPIN on the Solana blockchain. Nakajima embraced the community's momentum and began developing a formal project and framework around the Pippin persona. ^[14]​

Project Components

Pippin exists in several forms:

• AI Social Game: Nakajima developed Pippin into a fully on-chain social game built on Blast, an Ethereum Layer 2 solution. Described as an "AI stock picker game," it allows players and AI agents to compete in a simulated stock market. Players use "Pippin" NFTs, built on the ERC-6551 standard, which function as smart contract wallets to hold assets and build an on-chain identity. ^[2]
• Modular AI Framework: The Pippin Framework, released in December 2024, is an evolution of BabyAGI's concepts. It is a lightweight, modular architecture for building autonomous agents. Its key features include character definition (personas and goals), a dynamic memory system for learning and self-optimization, and seamless integration with external tools and APIs. ^{[4] [15]}
• Decentralized Creator Network: The project was also conceptualized as a "DePIN for creators" or a "pop-up startup studio." This vision involves a global network where creators can collaborate on projects and tasks, or "quests," without traditional corporate structures. ^[7]

The name and image of Pippin were released under a CC0 (public domain) license to encourage community-driven content creation and development. ^[4]​

Other Projects

Living up to his "build-in-public" philosophy, Nakajima has developed and shared numerous other experimental projects, including:

• askanything.vc: A search engine for insights from venture capitalists. ^[13]
• Mini-Yohei: An early AI experiment from 2022 to create an AI version of himself capable of writing blog posts. ^[13]
• Self-Operating Computer Framework: A framework enabling multi-modal AI models to operate a computer's graphical user interface. ^[4]
• Ditto: A minimalist, self-building coding agent that can generate a multi-file application from a single prompt. ^[11]

Public Recognition

Nakajima's work at the intersection of AI and venture capital has led to several public speaking engagements and recognitions. He was invited to speak at the inaugural TED AI event in San Francisco about his work on BabyAGI and presented on "How to run your VC firm using AI" at an Andreessen Horowitz (a16z) event during SF Tech Week. In 2023, he was named a Top 20 thought leader in the Data Driven VC Landscape report for his innovative work at Untapped Capital. ^[1]​

Interviews

Building Self-Improving Agents: Yohei Nakajima on “Every” #01

On October 23, 2024, Yohei Nakajima appeared on the Every YouTube channel to discuss the development of autonomous artificial intelligence agents and the creation of Baby AGI. During the conversation, Nakajima outlined the origins of the project and its role in introducing the concept of task-oriented AI systems capable of operating through large language models (LLMs).

He described how successive versions of the framework, including Baby AGI 2, were designed to enhance automation through iterative tool generation and refinement. Nakajima also introduced Ditto, an AI-based coding agent developed to produce multi-file software applications from text prompts, demonstrating an approach to program synthesis supported by advanced LLMs such as GPT-4 and Claude.

Throughout the interview, Nakajima discussed how autonomous agents might function as extensions of human workflows, assisting with distributed tasks and data organization. He drew a comparison between the development of AI agents and human learning processes, suggesting that knowledge representation through graph structures could improve how AI systems model relationships between concepts.

The discussion also addressed potential directions for future agent design, including systems capable of storing and reusing generated tools to create shared libraries of functionality. Nakajima emphasized modularity and adaptability as central elements in the design of AI systems and noted that similar principles inform his approach to investment in early-stage artificial intelligence projects.

The interview concluded with remarks on the continuing evolution of large language models and their potential integration with tool-use and reasoning capabilities, outlining a perspective on AI development centered on iterative improvement and autonomous learning. ^[16] ​

Yohei Nakajima’s Approach to Venture Capital and AI Integration #02

On September 25, 2025, Yohei Nakajima, founding partner of Untapped VC, participated in an interview on the YouTube program Carta hosted by Peter Walker. The discussion focused on Nakajima’s approach to venture capital, particularly his use of automation and artificial intelligence in investment analysis and operational processes.

Nakajima described his investment framework as a top-down approach, in which market trends are analyzed before engaging with founders. This contrasts with the conventional venture capital model, which often relies on personal networks and founder referrals. He indicated that social platforms such as X (formerly Twitter) and online AI communities like Baby AGI are tools he uses to observe developments in technology and identify early-stage opportunities.

During the interview, Nakajima explained how AI systems and autonomous agents are incorporated into his daily work to improve the efficiency of tasks such as deal tracking and founder communication. Despite this integration, he noted that personal interaction and human judgment remain essential elements of the investment process, as certain qualitative factors cannot be replicated by automated systems.

Nakajima referred to automation as a mechanism for expanding analytical capacity rather than substituting human reasoning. He mentioned that his workflow intentionally excludes fully automated evaluations or public communication systems to maintain consistency and reliability in decision-making.

The conversation also addressed broader developments in the startup ecosystem. Nakajima observed that the increasing number of AI-focused companies requires a more structured and disciplined approach to portfolio management. He emphasized the role of strategic alignment and thesis adherence over following temporary market trends.

In discussing geography and professional networks, Nakajima compared the ecosystems of Seattle and San Francisco, indicating that active digital participation enables access to broader technology communities beyond physical location.

He concluded by outlining a potential model for collaborative venture structures, where smaller funds operate under shared infrastructure. This approach, according to Nakajima, could improve coordination among early-stage investors and promote transparency in fund management. His publicly documented experiments and open development practices serve as examples of how transparency can be integrated into venture operations. ^[17]​

​