Where 9 Startups Are Building AI Apps, And $100 Million For Scientists
Upstarts speaks with Betaworks and founders from its latest AI Camp batch of startups emerging from stealth. Also: why Lux Capital's Josh Wolfe is committing millions to riskier science.
Upstarts is on the road again this morning, as I fly back to San Francisco for our first Subscriber Meetup co-hosted with Brex tonight, and to interview Plaid COO Eric Sager onstage at the NYSE Tech Summit later this week. I’ll also be demoing some seriously cool new technology and meeting with startup founders and funders around town, so stay tuned.
First, let’s turn our eyes back to New York, where a fresh batch of AI startups are hatching out of startup accelerator and investor Betaworks, and where another prominent investment firm, Lux Capital, is taking a stand on backing science.
Today is the demo day for nine startups in Betaworks’ AI Camp, a three-month residency for young companies building AI applications. AI is still the buzziest startup category to be playing in, and entrepreneurs have a number of accelerators, residencies, hacker houses and other programs eager to help launch them in exchange for some shares in their businesses.
We wrote about one such program, run by two longtime software design experts at Character Capital, two weeks ago. It would be a full-time job – and not a particularly valuable one for you, the Upstarts community – if we went about covering every one.
But Betaworks is an early player in the category, long before generative AI took off; one of its ‘camp’ residencies was where Hugging Face, the AI models and tools provider valued at $4.5 billion, got its start. More recently, AI note-taker and Upstarts community favorite Granola, which we profiled in April, worked with Betaworks early on.
So, given the sheer volume of promising AI startups already in the market, and the momentum of its leading players, Upstarts was curious to check in with Betaworks’ partner Jordan Crook and camp director Analisa Svehuag, as well as several of its current participants, to hear first-hand where they see opportunities remaining for new entrants in the AI race.
AI is also changing the expectations of what a company can achieve in a 13-week program, too. “You should be able to prove your thesis faster,” Crook tells Upstarts. “We don’t need all the bells and whistles.”
Last Friday, four other New York VC firms – BoxGroup, Union Square Ventures (USV), Lux Capital and First Round – co-hosted a one-day conference for a few hundred early-stage founders and soon-to-be ones. Aside from a few mentions on social media, the event operated mostly behind the scenes. (I attended in my capacity as Upstarts founder, gaining valuable feedback from entrepreneurs about what kind of news coverage they’re looking for.)
As the day came to a close, Lux publicly announced a $100 million-plus commitment to help scientists build startups or conduct research as part of what it’s calling a ‘Science Helpline’. Upstarts spoke with firm cofounder Josh Wolfe about that push.
“We want scientists to know there’s a different path,” says Wolfe. More below.
Presented by our Launch Partner, Brex.
Building a startup is hard. Managing your finances shouldn’t be. Brex gives you the speed, control, and scale to grow, without slowing you down.
✅ Same-hour access to high-yield banking
✅ Up to 20x higher card limits with bonus rewards
✅ Automated expense processing and invoicing
More than 30,000 companies—including 1 in 3 U.S. startups, such as Anthropic, Substack, Scale AI, and Robinhood—trust Brex to help them spend smarter and move faster. Run your startup on Brex → brex.com/upstarts
Opps In AI Apps
Two years ago, AI “wrappers” – applications providing a new or more convenient user interface on top of frontier large language models – were stigmatized. “We were so worried about what OpenAI, Google and others would productize,” says Betaworks’ Crook, who joined in 2022 after 12 years covering startups at TechCrunch.
Today, similar advantages that helped a wave of startups win in the rise of mobile apps are playing out, Crook argues: a short time to prove value, and network effects from popularity with early users, that can create proprietary data to improve a product in turn. “There’s always a risk that a big player or an existing, scaling startup tries to eat your lunch,” she says. “But it’s a race to value; to [find] who is in love with your product.”
As the ecosystem around the AI giants develops, the nine startups in this latest Betaworks batch skew more towards selling to businesses than the consumer-minded ones of last year. They’re also able to take advantage of ‘vibe coding’ – the use of AI to speed up software development by having it write, or assist, in generating code – to ship their own tools faster.
Several of the startups in Betaworks’ batch are building tools to bolster a sub-category of those popular tools known as an integrated development environment, or IDE. If you hear a developer talking about VSCode, or more recently Cursor and Windsurf (reportedly in the process of being acquired by OpenAI for $3 billion), those are IDEs, the latter two locked in a race for popularity among today’s AI-minded technorati.
Startup Decode is building a whiteboarding app that connects to such tools to allow developers to take a more visual-heavy approach to development. Hopper looks to make them more useful by bringing AI to bear on the pre-coding process: the product specifications and approvals process that plays out in companies before their engineers write any code.
Founder Nathan Ackerman, a former head of engineering at other startups, says that internally, the Hopper team builds its own product each day with Hopper on one screen, Cursor on another.
“Vibe coding is awesome. We do it every day,” Ackerman tells Upstarts. “But if anything, it reminds us that the hardest part has always been what to code.”
TabTabTab is a six-month-old startup looking to make AI models more useful by solving for their context limitations – automating the current manual process of copy-pasting a bunch of text from websites to a tool like ChatGPT and back.
Other uses of TabTabTab could include pulling company info from LinkedIn to put in a spreadsheet for a job search, or to make sure that AI-generated communications across a few apps, like email and iMessage, use a consistent tone.
Eventually, the startup hopes to be the central “orchestrator” with which a user would communicate with a number of AI agents carrying out complex tasks like managing travel, says CEO Gyanendra Mishra, who co-founded the startup with Vasin Wongrassamee in London.
A few of the startups are tackling specialized pain points within a business, like Trampoline, which is using AI to automate how service businesses fill out the Requests for Proposal, or RFPs, needed for them to sign up new customers to projects; JigsawML is attempting to use AI to manage a company’s enterprise software architecture more efficiently; Superposition is working to automate the recruitment process.
CEO Edmund Cuthbert spent seven years as a recruiter, helping to place Brex’s first engineer, before learning to code and becoming a startup product manager. His goal now is to provide AI software that eliminates the need for companies to hire recruiting managers to match with interested suitable talent.
“There doesn’t need to be a third human inserting themselves with little context and perverse incentives,” he says. Startups test-using Superposition have already made several founding engineer hires using the tool, which Cuthbert and co-founder Xiang Li spun up in the past 11 weeks.
Superposition’s secret sauce so far: a voice agent to which the hiring company explains their business and needs that can identify ideal traits in candidates the hirer didn’t know to seek. A food tech company, for example, might realize it needs food-loving engineers, identified because of their previous experience, or side projects they posted to GitHub. “That also unlocks a tier of talent who don’t reply to generic recruiter messages,” Cuthbert claims.
The other startups in the cohort are taking on specific verticals. Afterimage wants to help patients see what went wrong with medical procedures in the event of a complication, to better pursue claims or “justice.” NetAssist is building voice and text agents to help senior citizens and caregivers make more use of AI tools. And Graze Social is working on ways to monetize custom social media feeds built on top of Bluesky’s open protocol.
One challenge they all face that early mobile app makers didn’t: the lack of a centralized App Store for reaching early users. “Distribution is something we think about every single day,” says Crook.
‘A coalition of the confident’
Josh Wolfe is fired up. He’s taken a break from Lux Capital’s Monday partner meeting to talk to Upstarts about the firm’s new push to back science, and it sounds like something of a stump speech: American scientific leadership is under threat, the venture capitalist says. We’re in a race for scientific supremacy. China plays the long game.
What’s got Wolfe so passionate is a confluence of three factors: China leading the U.S. in an increasing number of research areas, the loss of domestic research funding, and an increasingly bureaucratic and politicized scientific environment, he explains. Add it all up, and you’ve got what Lux’s partnership considered enough of a crisis to want to step in and pick up some of the tab.
In October, Lux announced Lux Labs, an initiative for backing riskier science-based startup ideas. On Friday, the firm has expanded that effort to commit at least $100 million to support U.S. scientists through the Lux Science Helpline. Over its history, Lux has typically ended up investing 10% to 15% of its funds into research-driven, higher-risk but higher-impact projects. Lux Labs, and now the Helpline, ramp up those efforts.
Science startups typically only raise venture capital when they’re close to commercialization – maybe more than 90% of the way there, says Wolfe. “The science risk we’re willing to take with this initiative is a lot more, something like 50%,” he adds.
Lux and its partners are essentially willing to underwrite several million dollars of research that, until funding cuts under the current administration at the National Institutes of Health and other institutions like universities, might have been funded by grants and labs. If such research leads to venture-backed startups, Lux would benefit by investing, Wolfe says; but should scientists opt against launching companies, Lux could still connect them with limited partners who might want to donate more funding toward areas like medical breakthroughs, or with companies who might want to license technology or otherwise partner with them.
Lux isn’t trying to corner such a market as an arbitrage, he maintains, but welcomes other firms to join the initiative. “We want to help galvanize a syndicate of science-minded venture folks that will support this, because this isn’t something we can do alone.”