In 2016, two French founders landed in New York with an emoji chatbot for youngsters. They joined the primary program Betaworks ever ran for startups, BotCamp, and left with a $200K verify. The corporate was Hugging Face. This September, NVIDIA agreed to amass it for $12.93 billion, the most important tech acquisition in New York Metropolis historical past, for a platform that now serves greater than 18 million builders and hosts over 3 million fashions. Betaworks has been putting early bets like that for practically 20 years.
John Borthwick, a former Time Warner government and Fotolog CEO, based Betaworks in 2007 as a startup studio that constructed merchandise fairly than simply funding them. The studio produced Bitly, TweetDeck, Chartbeat, Dots, and Giphy. Twitter purchased TweetDeck. Take-Two purchased Dots. Meta purchased Giphy. Betaworks additionally acquired and relaunched Digg and Instapaper, and backed Twitter, Tumblr, Kickstarter, Medium, Anchor, and Gimlet alongside the way in which. It has since turned that intuition towards AI, with current investments together with Hugging Face, The Browser Firm, and Granola.
Camp, launched in 2016, is how Betaworks does it now. Every 12-week, in-person program within the Meatpacking District is constructed round a single thesis, with as much as $500K invested in each firm, break up between Betaworks Ventures and syndicate companions, and a Demo Day on the finish. Hugging Face CEO Clement Delangue nonetheless factors to it: “We began Huggingface at a Betaworks Camp so couldn’t suggest this sufficient!” Previous themes ran from bots and artificial media to App Layer, Interfaces and, this spring, Agent Methods, which drew greater than 400 candidates and picked eleven. In July 2025, Betaworks closed its $66M Fund III, aimed toward early-stage AI startups, with plans for not less than 25 pre-seed and seed investments plus not less than 50 by means of Camp.
This fall’s thesis was the brand new agentic financial system: what occurs to an organization when intelligence is not the bottleneck. Betaworks mentioned it wished to place “loads of chips down” on that concept. The Camp started September 14, with themes together with AI-native organizations, human and machine reliability, non-human marketplaces, and the bodily world, and it culminated on Wednesday, November 18, when eight firms took the stage at Demo Day and got here out of stealth. They spanned agentic commerce, AI staff for industrial corporations, an AI household workplace, coaching environments for brokers, voice, bodily AI for the trades, and warehouse imaginative and prescient.
“Eight firms every centered on the rising agentic financial system. Every accelerator cohort is framed by an rising know-how and a query we imagine will matter rather a lot sooner or later,” mentioned Borthwick. “This program is what occurs when intelligence isn’t the bottleneck, and you may construct a local AI firm. This isn’t in regards to the hole available in the market; it’s about constructing markets within the gaps, and that’s the place these eight firms are constructing.”
“I’m shocked at how restrained this group has been,” mentioned Betaworks associate Jordan Criminal. “There’s a quiet drawback round founders constructing every part as a result of they’ll construct something, and that’s not efficient in a crowded market. Every of those groups have gone deep on a selected drawback and their early traction is proof that it’s working.”
HERE ARE THE 8 STARTUPS FROM BETAWORKS’ AGENTIC ECONOMY CAMP:
Based by Bo Brainerd and Jorge Moreno, Ophelia is an API that lets any product with an LLM search, guide, and purchase, with each final result confirmed and recorded. It handles the authorization, navigation, transactions, and reliability that brokers want to show a request right into a accomplished motion. Its prospects are shopper software program firms constructing AI-native merchandise, from assistants and agent platforms to marketplaces and fintechs. The corporate says greater than 5,000 requests have already been accomplished by means of its platform.
Brainerd, the CEO, labored cybersecurity and operations for 12 senatorial and presidential campaigns, on the White Home and the DNC. Moreno, the CTO, constructed and scaled the ride-share group at Uber, main API infrastructure improvement for 40 million US prospects.

Intrinsic Labs turns an industrial firm’s working data into AI staff that do the work. The brokers run inside present techniques reminiscent of ERPs, TMS platforms, and CRMs. Ahead-deployed engineers provide the expertise these firms lack, then deploy, run, and enhance the employees for them. Prospects are industrial companies, manufacturing, and building firms doing $50M to $1B in income.
Cofounder Jon Slemp is a second-time founder with a previous exit and expertise working tech-enabled companies companies. Cofounder Yon Raz-Fridman, who joined in April, is a serial entrepreneur with a number of exits throughout {hardware} and software program. Intrinsic’s website cites outcomes from prospects together with a 90% quicker accounts-payable cycle at Colliers Worldwide and roughly 85% of handbook entry eliminated at Metallic Conversions.
X1 Wealth is an AI household workplace for households whose funds have change into a second job. It pulls a household’s accounts, paperwork, and entities into one file the household owns, prepares upcoming work reminiscent of capital calls, Ok-1s, and deadlines, and acts solely when the household approves, right down to making ready a financial institution switch. It really works alongside the household’s CPA, lawyer, and advisor, and connects to Claude and ChatGPT. The goal buyer is the household with a enterprise, actual property, trusts, or fairness compensation, and the professionals who serve them.
Cofounder and CEO Mat Silverstein spent ten years working development, product, and know-how at a agency that coordinated family-office work for entrepreneurs by hand. Jex Musa is cofounder. X1 presents a free tier, and expenses $97 a month for membership and from $499 a month for skilled workspaces. Creator Garrett Gunderson constructed his Multiplier Household Workplace on the platform.

Blobfish builds coaching gyms for AI brokers: simulated enterprise environments, with mock CRMs, ERPs, and spreadsheets, the place fashions follow actual workflows and are scored by deterministic verifiers. Consider it as a flight simulator for an AI worker. The pitch is specialised fashions which might be cheaper, extra correct, and in a position to run on-prem. The corporate says it has generated 2,000 domain-specific environments and used them to coach a 14-billion-parameter mannequin that beat a a lot bigger DeepSeek mannequin on the MCP benchmark, an early inner end result nonetheless awaiting broader analysis.
CEO Samuel Chien constructed RL environments for xAI’s Grok Enterprise Put up-Coaching group. Johnny Chien is CTO. Blobfish says post-training groups at Google DeepMind, SpaceXAI, and Alibaba, together with Mecka’s RL surroundings group, are reviewing the duties it generated and giving suggestions.

Heard is the voice context layer for brokers. As every of us finally ends up working with dozens of brokers, the thesis goes, we want a voice that is aware of what issues to us and what doesn’t. The corporate is constructing what it calls the subsequent voice interface for a billion customers. Its website describes full-duplex multimodal fashions that pay attention whereas they communicate and see what the person sees, with early fashions in coaching or preview and integrations with Google Calendar, Slack, Gmail, Notion, GitHub, and Linear.
CEO Kelly Solar is a two-time founder, a former HCI researcher and a veteran of Deloitte and McKinsey.
yaven builds AI teammates for firms of 1. The Mac app lives within the notch or a keystroke away from any app, works out what an impartial is making an attempt to do, finds the subsequent shopper and runs the work after the sure. It targets artistic, design, advertising, and PR companies, plus the freelancers and small studios turning into one-person firms. Context stays on the person’s Mac, it connects to twenty+ instruments, and nothing reaches a shopper with out approval.
Cofounder and CEO Bettina Sosa is a design engineer from Imperial whose work has been exhibited on the Design Museum. She holds two Artistic Conscience awards, has been a founding engineer twice and ran her personal company. Cofounder and CTO Nick Value went from Imperial to MIT, the place he wrote a prize-winning thesis, deployed embedded AI at an early-stage startup, and constructed AI infrastructure for a hedge fund.

Embarcadero Labs is constructing agentic bodily AI for frontline staff. Its product, Ten4, is an iOS app and cloud platform that works by means of a technician’s cellphone, Bluetooth mic, or sensible glasses throughout the job. It weighs video, voice, and images towards the client’s historical past, fills within the firm’s personal reviews and types, and hyperlinks every reply again to the second it got here from. Each job feeds a data graph of consumers, websites, and tools, so the subsequent technician begins with what the final one discovered. Early prospects are HVAC, electrical, and different high-turnover trades servicing information facilities.
CEO Lucas Cunha spent years implementing AI at Fortune 500 firms. CTO Dennis Nennos labored on early AR software program on the world’s largest chemical producer and researched computational physics at Harvard. The corporate’s website cites a 2026 Meta AI Glasses Affect Grant.

Pakket is warehouse automation constructed on giant imaginative and prescient fashions. It turns a warehouse’s present cameras right into a camera-based real-time location system, with no new {hardware}, giving reside visibility into forklifts, docks, and dwell time, plus a reside flooring map and KPIs tied to monetary efficiency. The pitch: most warehouses run at round 65% effectivity as a result of they’ll’t see their very own operations, which breeds errors and edge circumstances.
The founding group is CEO Dalton Dwyer, CTO Danick Desjardines, and CCO Jeff Huerta. Collectively they’re three-time founders with one exit and ten years in warehouse robotics.


