Venture Capitalist Reid Hoffman Questions Performance of Pure AI Funds
Reid Hoffman initiated a poll on social media, asking how a venture capital fund composed entirely of AI agents will perform in comparison to a fund made up of a mix of humans and AI in 10 years.
The poll offers three options: underperform, perform similarly, and outperform. At the time of the screenshot, there were 465 votes, with about 20 hours remaining, and the post has garnered thousands of views and dozens of replies.
Hoffman is a co-founder of LinkedIn, a partner at Greylock, a member of Microsoft's board, and involved with Manas AI (which uses large models to discover cancer therapies) and author of "Superagency." He has long advocated for the concept of AI agents, believing that professionals will shift from individual contributors to agent managers.
Some replies pointed out that the primary value of venture capital comes from human networks (clients, talent, follow-on capital), which are difficult for pure AI to replicate completely, suggesting that pure AI funds may underperform. Other replies believe that pure AI funds could outperform in much less than 10 years.
Currently, global venture capital has invested billions of dollars into AI agent-related companies, with institutions like Greylock continuing to invest in early-stage AI startups, including customer service agents, development tools, and scientific applications.
In terms of market mechanisms, traditional venture capital firms are directing funds toward AI agent startups, creating event-driven capital concentration; if pure AI funds are validated as effective, it will accelerate the shift of funds from relationship-dependent funds to algorithm-driven models, benefiting AI infrastructure providers while putting pressure on partners reliant on interpersonal networks.
Source: Public Information
ABAB AI Insight
Hoffman co-founded LinkedIn in 2003, became a partner at Greylock in 2009, made early angel investments in companies like Facebook, and promoted the concept of blitzscaling. In 2023, he reduced his responsibilities as a general partner in Greylock's new funds to focus more on AI projects, including seed rounds for Inflection AI and Manas AI, the latter focusing on accelerating cancer target discovery using large models.
His capital path continues to target AI companies with network effects or platform potential, leveraging his personal network to bring clients, talent, and follow-on financing to the invested companies. The motivation is to upgrade AI from a tool to an agent, partially automating capital allocation while retaining human judgment as the final decision layer.
This is similar to the process of quantitative funds replacing traditional subjective trading: early Renaissance Technologies surpassed humans in high-frequency domains using algorithms, but macro and relationship-based trading still rely on human networks. Currently, AI agents are penetrating from the execution layer to the decision-making layer, with the industry in an expansion phase, and pure automated funds have yet to be validated at scale.
Essentially, this is a combination of technological substitution and industrial chain reconstruction: AI can compress due diligence and screening costs, but LP relationships, founder trust, and cross-period capital coordination still depend on non-algorithmic interpersonal pathways. When agents can independently complete a full investment loop, pricing power will shift from individual partners to models and data providers.
ABAB News · Cognitive Laws
Networks determine access, algorithms determine speed
Pure automation is efficient, mixed decision-making is risk-averse
Capital first selects people, then is rewritten by algorithms