Basis Set Ventures closes $250M Fund IV in a single day, betting on generalized AI agents and physical world automation

Summary

Basis Set Ventures closed its $250 million Fund IV in a single day, a speed that founder Lan attributes to eight years of consistent positioning as an AI-first fund before the category became consensus. The Wall Street Journal broke the news the same day. LPs cited disciplined, early conviction as the reason for fast committee approval.

Basis Set launched in 2017, making it among the earliest dedicated AI venture funds. Lan's background is unusual for a VC: she began as a neuroscience researcher studying frontal lobe function and language acquisition in machines, then moved to Dropbox to learn how capital and growth mechanics work. Her first personal AI investment, pre-fund, was Scale AI.

Thesis and Portfolio Direction

The fund's core argument is that current foundation models have structural deficits. They lack persistent memory, hypothesis formation, iterative action loops, and sensory context, capabilities humans use automatically. Basis Set is investing across three layers to close those gaps:

• Infrastructure that makes models more capable, including portfolio company Mem Zero, which appeared on the show previously to announce funding
• Generalized computer-use agents, represented by portfolio company Similar, which automates work at the OS level across multiple underlying models rather than being locked to one lab's stack
• Physical world automation, including supply chain and construction, which Lan frames as a longer-duration bet given the difficulty of training models on force, temperature, and other tactile data

Lan is explicitly more bullish on system-level automation than browser-based agents. Browser-native agents are too narrow in her view; the higher-value outcome is full computer-use control where the user interacts only by voice or text and the machine handles execution entirely.

Physical World as a Long-Term Bet

Labor shortages in skilled trades underpin Basis Set's interest in physical automation. Lan cites welding as an example, noting welders can earn $3 million to $4.5 million over a career but that younger workers are declining to enter the trade. Robotics and construction automation remain earlier-stage bets because the data required to train physical-world models is harder to collect and standardize than software-side data.

Fund Parameters

Basis Set writes initial checks of $2 million to $3 million, with capacity to go up to $10 million at the initial stage. The fund targets two founder profiles: those working on technically difficult problems where the underlying technology does not yet exist, and those in competitive markets who can out-iterate incumbents. Lan describes the fund's role as being the first institutional believer before a market thesis becomes widely legible.