Betaworks doesn’t think AI agents are going to free up anyone’s time. The firm’s bet is that agents will take over more of the work and create a fresh wave of technical and human problems along the way, and it wants to fund whoever solves them first.
“The new agentic economy, as it emerges, is going to create a whole mess of second-order problems and challenges, and that’s actually where we’re spending our time and placing our bets,” Jordan Crook, general partner at Betaworks, said in an interview.
Betaworks closed its Fund 3.0, its largest to date, in 2024 and began deploying immediately, with a final close of $66 million in early 2025, though the raise wasn’t formally announced until that summer.. The fund runs two tracks, about 25 direct pre-seed and seed checks a cycle, and roughly 60 companies a year through its Camp accelerator, where startups get up to $500,000 tied to a single technology cohort rather than a broad sector bet.
The fund ran its first agent-focused cohort in early 2024, well before agents turned into the obvious bet. It was also an early backer of Hugging Face, the open-source model hub, which Crook says lets the firm watch up close how fast open-weight models are catching up to closed frontier labs.
Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 over rising costs, unclear business value, and weak risk controls. A separate MIT study found 95% of generative AI pilots at large organizations produced no measurable profit-and-loss impact, even as individual workers say they use the tools constantly. She calls this “the gap between individual and organizational gains,” and it’s where she expects the next wave of investable problems to show up, in what it actually takes to redesign a job, a team, or a whole organization around AI instead of just bolting tools onto the old one.
When DeepSeek R1 launched on January 20, 2025 as a free, open-source reasoning model, it briefly outranked ChatGPT on Apple’s App Store and knocked billions of dollars off major U.S. tech stocks. That kind of shock, Crook says, is proof frontier labs can’t lean on architecture alone as a moat, since open-source developers can reverse-engineer it. The value keeps getting pushed up the stack toward agents and applications; that’s the layer Betaworks is positioned in.
Consumer AI venture funding hit a record $89 billion in 2025, according to PitchBook, though nearly 95% of that money went into later-stage rounds rather than new seed bets. Investors are crowding into a shrinking set of application winners while the model layer underneath keeps getting cheaper to replicate.
Crook says commoditization is changing what a startup moat even looks like. Betaworks backed the meeting-notes app Granola, which launched against venture-backed incumbents Otter, Fireflies, and Fathom, plus Zoom and Google Meet, both of which added transcription overnight. “Moats now have to be earned instead of engineered,” Crook said. User loyalty, not defensible code, is what protects a startup once any interface can be rebuilt in days, she said. She goes deeper on the idea in a piece on Granola’s moat.
The second-order opportunity map
The fund sorts its bets into four buckets. One axis is who feels the problem, the individual user or the organization. The other is what fixes it, technical infrastructure or human behavior.
| INDIVIDUAL | ORGANIZATIONAL | |
| TECHNICAL |
Personal agent tooling • Notetakers earning loyalty daily, like Granola • Prompt and workflow personalization • Consumer hardware feeding world models |
Agent infrastructure • Verification and trust layers for agent output • Agent marketplaces and payment rails • Rebuilding the web so agents can transact |
| HUMAN |
Mental load and self-knowledge • Attention and focus management tools • Deeper self-comprehension products beyond AI tarot • Personal data legibility, e.g. Google Takeout tools |
Org and workforce redesign • Upskilling and reskilling programs • Job and team re-architecture around agents • Closing the individual-to-org productivity gap |
Agentic AI isn’t one trade, and that’s the case Betaworks is making with the map. The infrastructure quadrant, verification layers, agent marketplaces, what Crook calls “reterraforming the internet” so agents can transact on it, looks like classic picks-and-shovels investing. The organizational quadrant looks nothing like traditional enterprise software. Betaworks is leaning into services that can finally scale, and into vertically integrated companies built to replace the entire stack. The same signal applies to founders: the model layer isn’t the safe bet anymore.
Betaworks is also putting more of Fund 3.0 into consumer hardware than it expected when the fund launched, on the bet that physical devices will supply real-world data that world models still lack. Crook thinks the winners of the next funding cycle will be whoever solves the problems agents are about to create, problems that don’t exist yet.