Robinhood Ventures Fund II, the business development company that began trading on the NYSE today under the ticker RVII, opened at $25 a share with a portfolio built almost entirely around AI. The fund holds stakes in roughly 79 startups through $250K SAFE notes, plus a single $1M position, and leans heavily on Y Combinator alumni. It arrives at a moment when venture is already running hot: US firms deployed $320B across startups last year, the second-highest total on record. For the AI industry, the more interesting story is not the debut itself but what the fund chose to buy. Read the holdings list and a clear thesis emerges: the next wave of AI value is being built by agents, and by the startups that support them.
Watch the RVII roadshow.
Agent infrastructure is the fund’s biggest bet
Roughly a quarter of the portfolio sits in companies building the plumbing that AI agents need to work. The largest holding is Shortwave Communications, maker of Tasklet, an agent platform that owns work end to end and the fund’s only $1M position. Agentic Fabriq builds the control plane for AI agents, Arga Labs provides real-world sandboxes for testing them, and Apex Flux handles complex scheduling. ReasonBlocks supplies the runtime layer that makes agents cheaper and more reliable, while Second Stage Labs builds the communication infrastructure those agents depend on. Relay Innovations, via its AgentPhone brand, sells phone numbers to AI agents, Terminal Use orchestrates background agents, and Speedtrain, which now goes by Talking Computers, builds AI systems that run AI infrastructure. The concentration is hard to read as coincidence: the managers are betting that the tooling layer around agents compounds faster than any single agent product.
Vertical AI moves into the front office
A second cluster applies agents directly to business workflows. LegalOS runs an AI-native immigration law practice, KelAI builds autonomous quants for funds and traders, Caretta applies real-time AI to sales calls, and Daymi, now branded Pollen, automates customer success. Autumn AI surfaces real-time prospect signals, Samora AI builds multilingual voice agents, Unilabs brings AI voice agents to on-prem telephony, and Maywood AI pairs finance compliance with proactive outreach. PantaCapital runs an AI-native commercial insurance brokerage, InstaAgent scales marketing across hundreds of personas, and Sarah AI, now Corvera, supplies the context layer for AI-native consumer brands. Sparkley, which has since rebranded to Robby, sends AI agents after home-services revenue, while Carnot AI, doing business as Jinba, automates enterprise workflows through chat. The pattern is consistent: nearly every name sells automation into a workflow that still runs on manual effort, from legal filings to insurance quotes.
Foundations and developer tools
The rest of the fund reads like a shopping list for the AI developer economy. JigsawStack sells deterministic-task AI models, Smol Machines ships self-contained VMs for faster software delivery, and Keyframe Labs builds lifelike AI avatars. MirageDoodle, operating as AutoSitu, offers an AI-native workspace for development plan reviews, and Workable Solutions, known as /dev/fast, is building AI-native code hosting. On the data side, Luel curates rights-cleared training data and InkVell, now Synthetic Sciences, develops foundation models for scientific research. Oxus AI automates internal audit workflows, Qomplement is building agentic ERP for supply chains, and Unifold provides multi-chain payment infrastructure. Voygr Tech sells place intelligence APIs, Veriad, now Supafax, is an AI assistant for email and calendar, and Lambda Systems is building payments infrastructure for voice agents. What these names share is a bet on the tools developers will need as agent deployments scale beyond demos.
What the debut signals
Taken together, the fund reads as a bet that AI’s next chapter belongs to agents, not just models. Nearly every holding either is an agent or sells to one, and the Y Combinator funnel means the fund is priced into startup formation at the earliest stage. The BDC structure matters for the industry as well: it gives retail investors a way to own venture-stage AI exposure without the accredited investor test, and it forces the fund to publish its NAV and holdings on a regular cadence. That transparency turns a venture portfolio into a public window on where seed-stage AI money is flowing. For founders, the fund adds a new kind of backer to the cap table; for investors, it makes the earliest stage legible in a way venture capital rarely has been. If the agent thesis holds, RVII may be remembered less as a fund launch and more as the moment retail capital formally arrived in the agent economy.
