The Upshot

Serval co-founders Jake Stauch and Alex McLeod credit much of their recent success back to their time at another startup, Verkada.

As the heads of product and engineering at the security unicorn, they worked closely together on a feedback loop: talk to customers, build for their pain points, figure out how to sell it more widely, then talk to more.

“It’s the kind of founder magic that leads to such a great end experience for the customer,” says Stauch, Serval’s CEO. “Why would you ever give that up?”

But Serval, which launched two years ago with software automating systems for IT teams, and which has now expanded into areas like HR and finance, is growing too fast for its founders to still have their hands in every loop.

The company raised $75 million in December at a $1 billion valuation after revenue grew 6x in three months. A team of just about 60 staff now work with customers including tech high-flyers like Notion, Perplexity, Together AI and Vercel, but also almonds cooperative Blue Diamond Growers and media conglomerate Fox.

So Serval has come up with a novel experiment for scaling founder-level care: a program that explicitly hires aspiring founders to work as forward deployed engineers.

Called Serval Start, the program is launching with a cohort of one dozen members, with four already committed, and eight reserved for net-new hires.

Cohort members will be thrown at the company’s most strategic and biggest accounts, with the expectation that they’ll be able to troubleshoot and hustle their way to maintaining Serval’s “vertical line” growth.

What they get in exchange:

• Accelerated equity vesting schedule with a six-month cliff (compared to typical one year)

• Early exercise rights at Serval’s valuation from when they start

• Direct access to Serval’s execs including its founders

• Time with its VCs including First Round, Redpoint and Sequoia

• Weekly and monthly meetups and other programming

The program will be led by Christine Kim, who joined Serval two months ago after spending five years investing at Greylock. Its success will be measured, she says, not just by how participants boost Serval’s business, but also by the startups they eventually leave to build.

“There’s a lot of programs targeted toward founders that already have an idea, or a company in mind,” Kim tells Upstarts. “We wanted to find something that would be even earlier than that.”

Upstarts recently celebrated our first birthday. To celebrate, we’re offering 25% off a one-year subscription for a limited time 🎉

In a frenetic startup environment where investors and longtime operators are launching new-look residencies and founder matching programs to try to launch – and invest even earlier – in new businesses boosted by AI’s new capabilities, Serval’s program stands out as arguably the first formal, in-house program of its kind.

It’s not without risk, from hiring pressure to team dynamics, and what could work for Serval may not apply to every startup. Still, the program is worth the wider ecosystem watching closely – if anything, to see what new startups emerge.

How Serval set up the program, thoughts from two of its early members, and the risks and implications for other startups are below.

A MESSAGE FROM OUR SPONSOR

Lowering costs, simplifying care

UnitedHealth Group is helping simplify how patients navigate care by investing in transparency tools that make it easier to locate nearby providers and compare costs.

Transparent pricing can save patients hundreds of dollars annually.

Learn more.

‘Extended bootcamp’

Serval’s CEO Stauch hopes that Serval Start can spare founders from living his own experience before his current company.

Coaching companies isn’t new for him: at Duke, he helped build out the university’s entrepreneurship club, later serving as a mentor alongside his spouse Tatiana Birgisson, also Serval’s COO. But Stauch spent seven years building his first startup, which focused on brain-controlled marketing research and then video games, before moving on to lead product at Verkada.

“I could have cut off a lot of time there,” Stauch says. “This is the program I wish existed then.”

The idea of such a program first came up in a board meeting. It got legs – and Kim as its champion – because Serval leadership and investors were grappling with a problem of how to serve overwhelming customer demand.

The answer, Stauch and co. believed, was an employee empowered to build new features at customer requests, take initiative and problem solve like Serval’s founders did when they led all sales.

That doesn’t mean any old forward deployed engineer, argues Stauch: “There’s been a proliferation of what’s called an FDE in tech right now, it’s everywhere,” he says. “But if you look under the hood, most of these folks are just implementing software. It’s a rebranded solutions engineer or implementation consultant.”

Some elements of the program already existed informally. Employees known to be interested in starting companies might already get introduced to Serval’s investors; the startup also already intentionally hired former founders who might want to try again.

Take 25% off a 1 year sub

That was the case with Spencer Lang, an internal selection for the upcoming cohort. After working at Microsoft for several years, Lang had joined a startup as a founding engineer and then co-founded a startup intended to help non-profits use video in direct mail. When his co-founder departed the project, Lang decided to join Serval last October as an FDE to retrench and build up his network.

“I was pretty explicit upfront with Jake that I would start another company, and he was super supportive of that,” Lang says.

Working closely alongside fellow hires over a sustained period of time can give you a better understanding of who might make for ideal co-founders than searching through Y Combinator’s Co-Founder Matching Platform, or even joining a residency like HF0 or South Park Commons, he argues.

“I can think of five people here that I would start a company with, and have extremely high confidence,” Lang adds.

High talent density is what attracted an early outside applicant selected to participate. Felicitas Binz spent nearly three years working in product at healthcare automation startup Luminai before leaving in November to pursue her own startup idea.

Like with Lang, Binz’s project with her planned co-founder friends didn’t work out, and no natural co-founders materialized in her network. Joining Serval, she decided, will allow her to meet other high-achieving peers, while “challenging” her own product knowledge at the cutting edge of how AI software is being deployed.

“A lot of stars need to be aligned before you can go full-time on a startup,” she says. “Serval is a place where I can be more strategic about my next step.”

‘Great takes time’

It seems likely that Serval will have no difficulty filling the eight remaining spots in its first batch.

That’s when the hard part starts, Kim knows. The bar for success includes Serval Start hiring even more exceptional talent than its typical standards, she says; those hires will then need to meaningfully grow the startup’s business, while also taking advantage of the program’s opportunities.

On the business side, Serval already has a customer pipeline bigger than it can handle for 2026, Kim says. Whether it can accelerate deployments and revenue will depend in large part on its FDEs, and specifically this Start squad.

At the same time, Serval will expect Start members to work nights and weekends on their entrepreneurial journeys – not regular company hours.

If the cohort is too mercenary, lagging on their day work to prioritize the networking side, and leaving as soon as their equity vests in six months, the program will fail.

Serval’s leaders say that’s not as big a concern as it might seem from the outside, because personal performance is tied closely to the company’s. “We’re still small enough where you can’t really hide, and you can’t really look good without doing something that’s helping our customers,” Stauch says.

Still, everyone leaving within six months, and Serval having to start over with those FDE roles and the batch, would be a mixed outcome. “It’s a risk that we have to be excited about,” says Kim.

Bill Trenchard, the First Round partner who invested early in both Verkada and Serval and is volunteering to work with the batch, says he’s “not too worried,” and expects many of the cohort to last up to the intended two years.

“Actually learning what great looks like takes time,” he argues. “And the best people will want to make a huge difference, wherever they are.”

Get 25% off a group subscription

What if those startups bring engineers or other staff along with them, plunging Serval into a brain drain? The startup’s founders left Verkada amicably, and the company invested in the new idea. Serval’s best case scenario is probably that the Start members pair off with each other, but that’s not a given.

“There’s a balance to be struck there, and I’m sure there’s a threshold where it’s less cool,” says Lang. “That’s a bridge to cross when we get to it.”

A new playbook?

If you’re a founder or startup staffer reading this far, we’re guessing it’s because you’re at least curious to explore whether a version of Serval Start could work elsewhere, too.

It’s obviously too early to know whether Serval’s experiment will work out. But its organizers note a few criteria that made it seem worthwhile for Serval, that can serve as a checklist for exploring at other startups.

Traits for a company to benefit from founder FDEs:

• Serval is high-growth: Enterprise and Fortune 500 sized accounts come in every week, Kim says, meaning there is constant demand for staff to jump on ever-bigger, more complex deals.

• But it’s not too big: With under 100 employees, Serval still operates at a scale in which FDEs can actively build and ship features for customers – something harder to do at bigger organizations like OpenAI.

• It’s enterprise-focused: Such a program only makes sense where FDEs are at a premium, and when aspiring founders want to build in B2B or enterprise themselves. It won’t make sense for a consumer app.

• Its product has wide reach: Serval’s automations touch upon a number of teams within a customer, from procurement and finance to legal and security, as well as product and engineering. Such exposure might be less valuable for FDEs at companies that only sell to one team, such as engineering or sales.

Companies that could naturally set up such a program, per Stauch: Ramp in fintech, or AI agent builders Decagon and Sierra.

There’s one more variable to consider: ability to attract premier talent. If the people coming in aren’t more skilled in a certain area than your status quo – meaning you’re not unlocking access to stars you couldn’t hire otherwise – it’s unlikely to be worth the trouble for the company and investors alike.

“Maintaining selectivity is going to be the main challenge,” says Kim. “We already push ourselves to look for the top tenth of a percent of engineering talent, and now we have to look for the top one-hundredth of a percent.”

The Upshot

Apple’s longtime former AI chief John Giannandrea is joining a science AI startup, CuspAI, to help build out its U.S. operations, Upstarts has learned.

The former boss for efforts including Apple Intelligence, robotics and Siri plans to work part-time with CuspAI, three sources say, leading the Cambridge, U.K.-based materials lab’s efforts to open a Bay Area office with heavyweight AI talent.

The advisor role doesn’t yet have a formal title attached to it, but represents Giannandrea’s first move back into the ecosystem since his news-making retirement from Apple two weeks ago.

Giannandrea and CuspAI both declined to comment.

The move comes as CuspAI, which uses AI to help customers discover new materials, is reported to be raising $200 million or more in funding at a valuation of more than $1 billion. CuspAI previously announced a $100 million Series A in September, on top of a $30 million Seed round in 2024, the year of its public launch.

The reported upcoming round aside, Upstarts understands that CuspAI had raised more than that — more than $200 million total already — through a previously unreported Series A extension. Investors include NEA, Singapore’s Temasek, Hoxton Ventures, Basis Set Ventures, Lightspeed and the venture arms of Nvidia and Samsung.

What, exactly is CuspAI? And why is Giannandrea joining it, of all startups?

We have some early answers — both from the perspective of the exec known to colleagues as “JG,” and from that of the startup looking to compete in a fast-developing area of AI research — below.

Our analysis on the move, including those answers, is for paid subscribers. Join them as part of our first birthday sale: 25% off a one-year subscription, for a limited time.

Take 25% off

Read more

In 2020, Blumenfeld-Gantz’s exposure to his own parents’s struggle navigating Medicare — faxes, phone calls, even a trip to the Social Security office — led him down a rabbit hole of whether someone could build a better way.

“I was just appalled at how absurd the entire ecosystem is,” he says. “And not just the technology, which is a big part of it. There are also just terrible incentives that make every broker push plans that pay them more.”

So he launched Chapter: a startup focused on connecting Americans to better Medicare plans, saving seniors thousands of dollars per year. Recently, Chapter reached $100 million in annual recurring revenue and a $3 billion valuation after receiving backing from Al Gore’s fund Generation Investment Management.

For Blumenfeld-Gantz, that’s proof that his mission is distinctly bipartisan. Founded and led in New York City, Chapter has deep GOP ties, from Blumenfeld-Gantz’s legacy at Palantir to Chapter’s co-founder Vivek Ramaswamy, who was just a billionaire businessman then, but has since run for president and is currently pursuing the GOP nomination for governor in Ohio. One of Chapter’s earliest investors was current Vice President J.D. Vance, when he was just a VC.

Chapter is the only startup we know of backed by vice presidents from both sides of the aisle.

“When people spend time in the government, they see how big of a problem and how complex these systems are,” its CEO says.

On The Upstarts Podcast, Blumenfeld-Gantz shares his journey from foreign policy studies to forward deployed engineer at Palantir; what he learned at the tech company and why it’s proven a startup breeding ground; how Chapter’s nationwide approach helped it reach scale faster; and why he proposes a twist on a typical startup metric: product-market-channel fit.

Plus, he shares his Upstart Moment: when last year’s government shutdown cut off Chapter’s data right at the opening of Medicare’s busiest enrollment window, and his team had one week to reverse-engineer their way out.

• Listen on Apple Podcasts

• Listen on Spotify

This season of The Upstarts Podcast is presented by Mercury: banking, redesigned from the ground up.

Chapters:
00:00 Introduction
01:04 Solving his parents’ problem
08:08 From foreign policy to Palantir
12:01 Why Palantir breeds so many startup founders
14:28 Getting started and managing VC no’s
20:24 Why product market channel fit matters
26:51 Cobi’s Upstart Moment: the shutdown
32:33 “One of the most hated people in the industry’
35:41 Tuning out noise and talent density
42:37 Up next: a wider financial platform for senior spend

Just looking for the highlights? As always, we break down three startup lessons from the episode for paid subscribers below.

Unlock all of our paid posts by taking advantage of our first birthday sale: 25% off an annual subscription, for a limited time.

1. Palantir tips: ‘unstructure,’ ‘meaningful problems,’ and ‘eating the pain’

2. Going national from day one

3. The value of ‘product-market-channel-fit’

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a big difference and will help the show continue and grow.

Read more

At an exclusive AI-focused event on Monday afternoon, Figma board director Andrew Reed asked CEO Dylan Field to talk about his relationship with AI lab Anthropic.

It was a loaded question. Just three days prior, Anthropic had announced Claude Design, a product that seemed to compete with Figma’s software. That move came only three days after Mike Krieger, co-lead of the same Labs team that developed it, had formally resigned from Figma’s board.

Towards Krieger, known for his role as co-founder of Instagram and until recently Anthropic’s chief product officer, Field was polite. “Mike is a great guy, and he was awesome on the board,” Field said, as confirmed by two sources who were in the room.

For Anthropic itself, Field chose more loaded words:

“They were not consistently candid in their communications,” he said.

This exclusive story is for premium Upstarts subscribers. Join them in our first birthday sale for 25% off a one year subscription for a limited time.

Read more

The Upshot

Some people are ‘inbox zero’ types, making a point to read – maybe even label or archive – every email that comes in each day.

Naveen Gavini is not one of those people. At least not when it comes to his personal life, where he calls his habits more like ‘inbox infinity.’

Until his new startup, Gavini worked as chief product officer at Pinterest, where email “ran his day” on the job. Meanwhile, in his personal inbox, other, arguably as important, notes would slip through: friends’ birthdays and major family updates, notes about his kids, even his pet’s insurance coverage lapsing with an expired credit card.

“Every time I’d open up my inbox, I felt this little anxiety of the never-ending to-do list,” Gavini says.

So after teaming up with two friends from Pinterest – Albert Pereta, its former executive creative director, and Steven Ramkumar, a former distinguished engineer – Gavini decided to take on a problem left largely unsolved since the launch of Gmail more than 20 years ago.

Their new app, Extra, leverages AI capabilities to make inboxes more personalized and proactive.

And after five months in beta testing with about 1,000 users, Extra publicly launches its availability today. Initially open to Gmail users via iOS and web, Extra will onboard people who sign up off its waitlist daily, as fast as they can manage, Gavini says.

(The first 100 or so Upstarts readers to sign up can skip the line with the code UPSTARTS – we’re choosing to share this, and are not being compensated in any way to do so.)

In a few days testing Extra at Upstarts, we saw first-hand what the team is trying to build: a new-look, more responsive, more fun inbox experience; albeit one that is still a work in progress.

If Gmail looks, as Gavini argues, like a spreadsheet of columns and rows of people and brands asking you for something, Extra presents a more colorful executive summary of what it thinks actually matters to you.

And while “email is the hook,” as Ramkumar puts it, there’s a bigger vision in play here: to build your “life inbox,” part personal assistant, part smart directory to the thousand little things going on in your life.

That’s why the startup behind Extra has a wider-reaching name, BuildForever; it also helps explain how the startup has already raised $9.5 million from investors including Felicis Ventures, Abstract Ventures, A* and Elad Gil.

BuildForever’s founders know the history here, too. Many startups have tried to tackle email since Gmail, from Mailbox to Superhuman; many have failed; the best have been acquired.

They’re making two bets provocative enough to deserve discussion by the wider ecosystem, even if you’re skeptical.

• Tools like Gmail weren’t designed for the LLM era

• AI builders are under-serving the consumer market

So much money and attention is focused on re-thinking our work tools and processes today, Gavini notes, with many of tech’s smartest people bringing AI to verticals like legal or healthcare.

“My workflow changes every three months now,” he says. “So why are the apps that I use on my phone every day still the same?”

Upstarts recently celebrated one year in operation, and we have big plans for Year Two. To receive our premium coverage and support our work, take 25% off an annual subscription for a limited time.

Support our work

Extra difficult

Email startup veterans have simple advice for the BuildForever team: don’t do it.

“We’ve learned a lot in the past year,” Gavini responds. “But as a consumer, I would love better personal email, so we’re going to take a shot.”

For such a potentially Quixotic venture, their credentials are strong. At Pinterest, they not only worked in consumer at scale, but launched Shuffles, a standalone collage app aimed at Gen Z that temporarily crested the U.S. App Store in 2022.

And the trio spent months chatting and researching the problem before diving into email eyes wide open. Their thesis: many of the more recent tools to emerge, like Superhuman, have focused on work, because that’s where the obvious money is. But how we use email for work is fundamentally different than when we’re off the clock – much more about continuous, repetitive back-and-forth or outreach; much less about passive consumption.

Busy professionals also tend to try to optimize their workdays first, then scrap together the rest. Not everyone was ‘inbox infinity’ like Gavini, but even ‘inbox zero’ purists like Pereta would spend hours a week tending to their manicured inbox gardens.

Then there are younger people, like the college students who tried Shuffles, who see Gmail as “for old people,” Gavini says. Many of them are now younger than Gmail itself; their names are already taken on the service.

Extra tries to solve the needs of all those groups, although it tends to resonate first with people like Gavini who want to ‘Marie Kondo’ their email backlogs.

At launch, Extra summarizes your daily meetings and what it thinks are key messages, like the delivery status of an Amazon package, or in my case, a looming deadline to purchase college reunion tickets.

For newsletters like Upstarts, Extra could be a double-edged sword. A handful show up in the front-page daily briefing, summarized to potentially drive more follow-through; the rest get grouped into one of Extra’s smart categories, which users can create and edit. Others might include a Kids one for parents, or Reservations for foodies.

Extra has leaned into smart actions and search. According to Pereta, too many apps still take the approach of a “blank box” – putting the pressure on a user to know what to ask, or retrieve. Emails in Extra suggest a few actions, like adding to calendar or going to a ticket page; you can also query a chatbot similar to productivity tools to ask for suggestions and find old context.

In chats with three test users identified by Extra, we heard three different reasons for adopting Extra. Emma Bates, a startup founder and fractional exec, finds it easier to toggle between six different email accounts without feeling overwhelmed. She’s set up a category for planning her wedding, too.

Take 25% off a 1 year sub

Emily Yi, a game designer, finds the interface more engaging and less “administrative” than using Gmail on her computer or the Mail app on her phone. As she helps manage care for two elderly parents, she values Extra’s voice capabilities: “The personal assistant part is huge.”

And Courtney Knights, who works in fundraising and development at Columbia University, has used Extra to help break a pattern of relying on two personal emails: one for important info, and another for marketing sludge. Extra has helped remind her to fill out a form for her kids’ summer camp; she’d like a way to connect multiple users, so she could drop updates around their school straight onto her husband’s page and calendar.

Our own experience with Extra has been more of a mixed bag so far. Little things like the reminders and package tracking are appreciated and immediately useful; so is having a simple, visually-pleasing central dashboard of personal tasks.

At other times, the app’s growing pains are more visible. One of the most amusing was when Extra made its first recommendation of a potential email source for me to unsubscribe, due to disuse: official emails coming from Extra itself.

Extra also appeared somewhat lost in time, recommending a year-old dinner reservation as needing action, and offering to check me in to a flight happening in two months.

Gavini says that’s partly the app showing a new user what might appear in certain categories; the app also seems to struggle a bit more with people like me, who take the time to label and archive email whenever possible.

As noted by the test users, it’s easier to get into the swing of using Extra on mobile. It’s also more valuable for people who already use multiple phones, or apps, for work and personal accounts, versus connecting everything into one Gmail inbox.

Racing ahead

How will Extra make money? It’s free for now, and to go up against existing tools, some version of it will likely need to stay free forever.

“We need to be able to support a billion people using the product, and I don’t think a billion people will pay $5 a month when they have a free service,” Gavini admits.

As it adds more assistant-like capabilities, Extra could explore a premium offering; it could also lean into marketing. You can already organize all of your emails from brands into one category to keep track of sales; would a brand gladly pay an affiliate fee to stay top of mind there? If Extra knew you were going on a trip, would Uber or Lyft pay to have a one-click option to book a car from the airport?

“Layering on that funnel is very interesting,” says Pereta. It’s also dangerous – similar to the tension of OpenAI exploring putting ads, or commerce buttons, directly into ChatGPT.

Get 25% off a group subscription

Gavini brainstorms that Extra has much of the info needed to track and identify rewards and special offers coming in from your various credit cards or loyalty programs. Would you, or the cards, pay to keep those top of mind and aggregated in one place? That opportunity reminds us of Paribus, the price-matching tool previously built by Ramp’s founders, and which was acquired by Capital One after being throttled by Amazon. The receipts are still in the email.

A more existential question facing Extra’s builders: if they’re right, and consumers want a new tool here, why won’t Anthropic and OpenAI simply gobble the market up?

It’s not a question that worries Peter Deng, who led the startup’s most recent round at Felicis, and who previously helped oversee ChatGPT at OpenAI. Besides BuildForever’s founders’ own experiences at Pinterest (the now cliche word “taste” comes up), he takes heart in the fact that the labs are focused on the revenue opportunity of work use cases: “They are currently in a world where the user talks first, and the model doesn’t talk at all,” he says.

From tools built DIY using OpenClaw, the recently popular open-source agent builder, to a hypothetical Claude chief of staff, it all might converge on the same super assistant opportunity. For consumers, more competition could mean better options.

“I’m not trying to convince other people to root for this company,” says Deng. “I just want people to try it.”

The Upshot

Trusting AI model output as the gospel truth is already a bad idea. It’s an even worse idea in women’s health, where Midi Health co-founder and CEO Joanna Strober is quick to point out that clinical research routinely excluded women as recently as 1992.

That’s a big problem when AI models are training off of, or drawing inferences from, research data before that cut-off to inform current-day insights.

“A lot of the research you see out there is just incorrect when it comes to women’s health,” Strober says. “A woman is not a small male. Unfortunately in most data models, they are.”

It’s one of a few reasons that Midi Health and Oura Health are both training their own AI models, alongside decreased funding at NIH for research more broadly. “It’s a systemic problem,” says Oura CEO Tom Hale.

Upstarts interviewed both health tech leaders on the main stage at HumanX last week. We also led a discussion on how AI is changing collaboration at software unicorns Gamma and Miro. Gamma co-founder Jon Noronha and Miro founder and CEO Andrey Khusid both had interesting points on how agents are transforming their businesses.

(Our other HumanX interview, on how DoorDash co-founder Stanley Tang set up its robotics division, is available for all subscribers here.)

For our paid subscribers, we’re breaking down the key takeaways from all four startup leaders below, covering AI’s impact in patient care, wearables, management and presentations.

“Everybody is becoming an executive,” says Khusid. “Most of us never wanted that life.”

As we continue to celebrate one year in operation, unlock this story and the rest of our premium coverage for a limited-time offer: 25% off a one-year subscription.

Read more

• Listen on Apple Podcasts

• Listen on Spotify

This season of The Upstarts Podcast is presented by Mercury: banking, redesigned from the ground up.

Chapters:
00:00 Introduction
01:11 Cohere’s ‘takeoff phase’
05:09 Where vibe coding falls short
08:05 Smalltown Canada to AI epicenter
14:27 Co-authoring the Transformer paper
18:52 Starting up Cohere
25:22 Why AI sovereignty matters
30:25 Aidan’s Upstart Moment
32:56 Playing on ‘hard mode’
36:55 Ethical questions of AI
39:44 Why Cohere should IPO

Just looking for the highlights? As always, we break down three startup lessons from the episode for paid subscribers below.

Unlock all of our paid posts by taking advantage of our first birthday sale: 25% off an annual subscription.

1. Be more like Jensen

2. Control > hyperscaler funds

3. IPOs are a good forcing function

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a big difference and will help the show continue and grow.

Read more

When DoorDash co-founder Stanley Tang pictures you ordering lunch, or a new phone charger, he doesn’t see a person driving it to you. He doesn’t even see a car.

Robotaxis were designed with transporting people in mind, not a sandwich.

“You’re just trying to retrofit that form factor into a delivery use case,” he argues. “If you think about it from first principles, why would you need a 4,000 pound car?”

That line of thinking helped lead Doordash to design the Dot, an electric autonomous delivery robot that it started publicly testing in Phoenix, Arizona last year. About four feet tall, the Dot can travel up to 20 mph, use sidewalks, bike lanes and roads, and carry up to 30 pounds on its own 350 pound frame – what the company bills as a safer and more efficient way to move goods.

When Tang, one of four now-billionaire founders of DoorDash, first set up its Labs unit in 2018, the company didn’t expect to be building its own robots. Instead, its autonomy strategy was focused on partnerships, Tang says.

We talked about that journey and lessons about setting up an R&D unit, as well as DoorDash’s thinking about autonomous vehicles and robotics more specifically, last week at the HumanX conference in San Francisco – on the main stage, as well as backstage for more to bring our Upstarts readers.

(Tang unsurprisingly didn’t give us a heads up that DoorDash would be participating in a controversial ‘delivery’ and press conference with the White House several days later; The Guardian can catch you up there if curious.)

You can find our key takeaways and video clips from the conversation below.

On Friday, we will publish more startup takeaways from our other interviews at HumanX:

• AI and data in health and wearables, and how women have been left behind, featuring Oura CEO Tom Hale and Midi Health CEO Joanna Strober

• How two productivity unicorns are thinking about AI in workplace productivity both internally and for their customers, featuring Gamma co-founder Jon Noronha and Miro CEO Andrey Khusid.

That post will be for paid subscribers only. If you’re curious what they had to say, or simply want to support our work here at Upstarts, it’s a great time to participate in our first birthday sale: 25% off an annual subscription.

That comes out to less than one of DoorDash’s sandwiches each month.

Partner vs. build

DoorDash initially intended for Labs to set up partnerships with other hardware companies for its autonomous strategy, not build its own.

DoorDash would offer APIs to connect to its software, handle operations and distribution, and leave the rest to other startups, Tang says.

In practice, things weren’t so simple. DoorDash’s surface area was broad and heterogeneous – urban areas, suburban, rural; short, medium, long distance – meaning few startup partners had more than a solution for one use case or persona.

Where hardware partnerships have ended up making the most sense so far: drones.

There, DoorDash partners with Wing, the delivery company owned by Google parent Alphabet that was itself a product of its ‘moonshot’ X Lab. (Co-founder Larry Page also launched and eventually wound down a flying car startup, Kittyhawk.)

DoorDash also works with Flytrex, an Israeli drone startup that operates in the U.S., and Manna Air Delivery, an Ireland-based startup.

(In 20223, I watched a Manna flight firsthand outside of Dublin with Stripe co-founder John Collison, one of its earliest backers, as part of a Forbes cover story on Collison and his brother Patrick.)

Worth the weight

Where DoorDash wants to operate now: the platform between merchants and customers that can connect any autonomous or human delivery modality to best fit the environment, distance and order type.

Drones will make sense for more remote or rural deliveries; robots like Dot for more dense suburban ones. DoorDash doesn’t have a good answer for high-density urban environments like the streets of New York City yet; most of its business happens in the suburbs anyway, says Tang.

DoorDash calls the system deciding its best delivery mode the Autonomous Delivery Platform, and it’s a selling point for Tang with potential partners and hires. “There’s a lot more to autonomy than just autonomy,” he says. “You’ve got to build the ecosystem around it.”

Doing so has led DoorDash to ship another product its founders didn’t envision, besides the Dot: a new version of an old-school scale, for weighing items.

The problem: drones have strict weight limitations, and DoorDash didn’t know how heavy the food orders it facilitated were. In order to work with Wing, the company had to design and send customers a scale to check them; then it started collecting that data.

“We started realizing, ‘hold on a second.’ This is actually super valuable data that not only works for drones, but could potentially work for robots, too,” Tang says.

Today, DoorDash has shipped more than 20,000 of its SmartScales to merchants. The scales have had another unforeseen benefit beyond autonomous delivery, too: the data helps restaurants know when orders might be missing items, because they’re coming in too light; participating customers have seen a 40% reduction there, says Tang.

Working backwards

As DoorDash expands its use cases into parcel delivery, Upstarts asked Tang if we should expect more versions of its Dot robot, or other robots that could specialize in packages of varying sizes.

Tang, the first DoorDash leader to confirm publicly that it’s currently testing autonomous package delivery, said it’s a fair guess. Test users have already ordered items like cameras and Mac Minis for setting up OpenClaw agents, he says.

Where lab, R&D or ‘innovation’ teams have failed inside scaling startups or recently public tech companies, Tang argues, is that they don’t work backwards from a business outcome, Tang argues. (The current wave of AI and robotics labs could take note, too.)

“A trap a lot of these kinds of research labs or hard tech labs fall into is they start with the technology first, build that in a vacuum, and then they try to figure out what use cases to try to fit it into later,” says Tang. “Whenever that happens, it always feels like the product isn’t quite the right fit.”

Startups or companies looking to design programs like DoorDash’s autonomous one should also ask themselves where the data sits, Tang adds. DoorDash had 10 billion lifetimes deliveries to use as data, and even then realized it was missing a key component with order weight.

It also needed data on a number of last-mile questions, like how to avoid parked cars, navigate roaming dogs, or reach a doorbell. “The hard part isn’t building a prototype,” Tang says. “How do you deploy the autonomy across multiple categories, at a global scale?”

To reinforce the point, Tang notes that DoorDash manages 9 million deliveries per day. A one-in-a-million chance of something happening will typically feel pretty remote. For DoorDash, that means it could be happening eight or nine times every day.

As DoorDash’s autonomous deliveries roll out, it’s easy to imagine a bunch of edge cases that could range from headache to fiasco. What if people realize its package delivery drones are carrying high-value items, and are easy to break into? What if more videos go viral of delivery robots getting stuck? What if one of its drones misses and puts a package through a sunroof?

Those are problems for Tang and partners to watch, but Tang believes that having the data and distribution is the most important place to start. Then DoorDash will keep building solutions for use cases where it doesn’t find strong options, startup or not.

His last word for now: “This is here already. It’s not a science fiction thing.”

When it comes to calling customer support, startup CEO Jesse Zhang knows your pain.  ”It’s just a very frustrating experience, because you’re stuck in a loop,” he says.

In 2023, he and co-founder Ashwin Sreenivas launched Decagon to solve the problem with powerful new technology: AI agents. Today, customers like Avis Budget Group, Duolingo and Fanatics all work with Decagon, helping it reach a $4.5 billion valuation.

But as AI unlocks new possibilities in support, Zhang faces serious competition from more established and better-funded rivals. We’ve written about this race and how Sierra, led by former Salesforce co-CEO and OpenAI chairman Bret Taylor, has leaned on enterprise ties to carve out its own space in the market, alongside others like Intercom and Forethought, recently acquired by Zendesk.

And on The Upstarts Podcast, Zhang was candid about that competitive battle, sharing how Decagon values speed and “go-to-market execution” as “advantages,” but not “long-term differentiators.”

He also shared his take on why OpenAI and Anthropic won’t move in on Decagon’s business; how he’s using two layers of agents to solve customer problems through a system called Duet; and why he envisions this category moving from support to other conversations through an AI concierge.

Plus, he shares his Upstart Moment: how he and the team scrambled to win an uphill sales battle by converting “red dot” skeptics into “green dot” buyers.

• Listen on Apple Podcasts

• Listen on Spotify

This season of The Upstarts Podcast is presented by Mercury.

Chapters:
00:00 Introduction
02:38 Why support was still a problem
05:34 Jesse Zhang’s founder story
15:23 Three pillars of startup strength
16:20 A 10x name and first customers
19:24 Winning with speed and scale
24:20 Jesse’s Upstart Moment in sales
31:35 Why you need to be 10x better
33:43 Revenge of the AI ‘wrapper’
37:17 Long-running agent ‘duets’
40:44 Your future concierge

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a big difference and will help the show continue and grow.

Just looking for the highlights? As always, we break down four top startup lessons from the episode for paid subscribers below. Unlock all of our paid posts by taking advantage of our first birthday sale: 25% off an annual subscription.

1. Three founder traits for a free-for-all

2. How to win hyper-competitive sales

3. How AI ‘wrappers’ can hold off the labs

4. Use agents to improve your agents

Read more

The Upshot

Yesterday, Anthropic announced a large-scale initiative to essentially wall off its own latest frontier AI model with yellow caution tape.

The unreleased model, called Claude Mythos Preview, is already so powerful a cybersecurity tool that it identified thousands of ‘high-severity’ vulnerabilities across every major operating system and web browser, including the one you’re probably using to read this article, Anthropic said.

In response, Anthropic announced Project Glasswing, an initiative looping in more than 40 companies like Amazon, Apple, Google, Microsoft, NVIDIA and JPMorgan Chase to early Mythos Preview access, along with $100 million in usage credits, so they could start to address these weak links.

For startups, that means business as usual, right? The coalition of the giants is on it. No need to worry about hypothetical issues now being addressed by some of the world’s biggest cybersecurity teams.

The reality: when it comes to cyber risks, Anthropic’s already-released tools, including the hyper-popular Claude Code, are plenty powerful as it is – as demonstrated by a new report out today from a cybersecurity startup, LayerX.

Anthropic isn’t the only AI lab with vulnerabilities being flagged by startups. But given the heightened popularity of Claude Code among startup builders and programmers today – a big reason Anthropic recently announced its revenue run rate has reached $30 billion – such findings, and Anthropic’s spotty responses to them, should serve as a warning for startups leaning on its tools.

“We’re not kicking the dog because it’s down,” says Roy Paz, the AI researcher behind the new report. “We’re kicking it because this is the gold right now, and Anthropic are the only ones truly capable of this [kind of attack]. Everyone wants this nugget of gold.”

Upstarts recently celebrated our first birthday. To read the rest of this post and support our work reporting on the startup ecosystem, take advantage of our best-ever offer: 25% off an annual subscription, available for a limited time.

Read more

Midway through our conversation, Axiom Math CEO Carina Hong casually slips in: “Math is AGI.”

The line obviously caught our attention, and hopefully yours, too. What’s she mean?

To hear Hong explain it, math and code are intertwined. Computation gives you an output; a mathematical proof gives you a property. Combine the two, and you can get real-time verification of code that works, and doesn’t collapse as spaghetti slop.

The business implications are wide-ranging: faster scientific discovery and research cycles; cheaper and safer testing for industrial applications that protect human lives; scalable ways to maintain regulatory and compliance requirements.

“The part that is actually quite beautiful is that it’s not about erasing hallucination or catching mistakes,” she says.

Instead, it’s about including the rigor of a mathematical proof in the process before they happen.

It’s a big enough opportunity that last year, Hong left an academic track, as a Morgan Prize winner at MIT, Rhodes Scholar at Oxford and joint law student and PhD candidate at Stanford, to start her own company.

In just nine months, Axiom Math shipped a tool that can solve some of math’s more advanced problems, and Axiom raised $200 million at a $1.6 billion valuation. But Hong, a first-time founder, is very much an Upstart looking to hold her own alongside the big AI labs and competitors like Harmonic, the unicorn co-founded by Robinhood CEO Vlad Tenev, and Logical Intelligence, whose launch we covered in Upstarts in September.

On this episode of The Upstarts Podcast, Hong shares her founder journey from immigrant at MIT to Oxford, and ultimately dropping out of Stanford; how she’s learning as a first-time founder to help Axiom compete in a red-hot new category; and her Upstart Moment when Axiom took the world’s hardest college-level math test, the Putnam Competition.

• Listen on Apple Podcasts

• Listen on Spotify

This season of The Upstarts Podcast is presented by Mercury.

Chapters:
00:00 Introduction
00:48 Intro to Carina Hong
02:02 What Axiom Math does
05:51 ‘Math is AGI’
08:48 Not replacing mathematicians
13:28 Winning the Morgan Prize
17:17 Origins of Axiom
22:22 The new math AI race
25:51 Carina’s Upstart Moment
31:47 Axiom’s business prospects
36:39 Why the future is verified coding

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a big difference and will help the show continue and grow.

For paid subscribers, we break down our top 3 takeaways from our chat with Carina below:

1. Verified code can be a better copilot

2. You still need humans to bring curiosity

3. Embrace a lifelong beginner’s mind

Read more

When Somos Internet CEO Forrest Heath first started hooking up homes in Medellín with better internet in 2018, even he wasn’t sure what where it was going.

Social project? Side hustle? Colombian adventure before moving back to the U.S.?

Whatever it was, Heath grew obsessed with the puzzle. Network limiting speeds? Architect a new one. Routers too slow? Make those, too. Software and firmware can’t keep up? Write better.

“It captivated me that it seemed like there was so much headroom to build better infrastructure,” Heath says. “It just felt like nobody had really thought through a lot of the underlying principles.”

In just a few years since, Heath – now effectively a Medellin local, with scruffy facial hair, signature rainboots and a Toyota Hilux – has built Somos into a high-growth, soon-to-be international business.

Today, Somos already employs more than 1,000 people and has become Medellín’s third-largest provider, the company says, while growing fast in its second market, Bogotá. Somos says it works with 80,000 customers overall, growing 6,000 each month; the startup projects revenue to reach about $30 million this year, then more than double in 2027.

Now, Somos is launching in Mexico City as its first international market outside of Colombia. The company is also working on technological improvements like a recently-announced fiber network inspired by data centers, called FiberX.

“We’re not just building another functional internet,” Heath says. “We’re building the internet that everyone wished they had.”

To build it all, Somos has raised $40 million in a Series B funding round co-led by Bracket Capital and Ribbit Capital, Upstarts exclusively reports. Union Square Ventures, Kaszek Ventures, Not Boring Capital, Zero Infinity Partners and Y Combinator participated.

Investors claim that just as a better telecom business for Colombia, Somos can grow into a multi-billion-dollar company. But what Somos so exciting – and why it’s interesting to Upstarts as a company to watch even if you have no Latin American ties – is Heath’s far greater ambition to build a new kind of infrastructure layer for the AI era.

Heath doesn’t just run Somos; he also founded a sister startup, Autoridad Panandina (APD), that’s looking to build data centers for AI computing needs close to Colombia’s ample hydroelectric power.

To celebrate our first birthday at Upstarts, we’re running our best sale ever on a premium subscription, for a limited time: 25% off an annual membership.

You’ll receive exclusive premium stories, get first word on events, and support more in-depth, original reporting about startups — all for less than the cost of a monthly sandwich.

The vision: a more closely connected Americas corridor in which someday U.S. companies can buy cheaper compute and power from as far as Colombia, or even Chile; and in which Somos and APD enable people to effectively run supercomputers in the cloud, leaning on effectively unlimited internet connectivity to run AI models and agents cheaper and faster than stringing together a bunch of Mac Minis.

But to do it, he’ll need to become one of the startup world’s best recruiters. He recently hired Jill Szuchmacher, a former Google Fiber leader, as chief operating officer; to scale like Heath wants, Somos will need to hire a lot more tech talent, and not just people with Latin American ties.

That makes Heath, who recently wore his boots to The Hill and Valley Forum in Washington, D.C., a potential partner – and rival for talent – with some of the startups in America’s emerging industrial startup corridor.

“I think in a really weird way, we’re this alternative bet on building crazy, ambitious infrastructure, not just in El Segundo, but doing it in places outside the U.S.,” Heath says.

More on how Somos is doing it, what customers and investors say, and lessons for other startup builders below.

A MESSAGE FROM OUR SPONSOR

Radically different banking loved by over 300,000 entrepreneurs

When I launched Upstarts last year, I knew we needed a banking partner that could move at our speed, and grow with us, too.

Mercury is thoughtfully designed for the unique needs of entrepreneurs like me, offering an easy and intuitive banking experience so I can spend more time with the founders we like to cover.

Building a media startup comes with new challenges every day. But with Mercury, banking is one we don’t need to worry about.

Visit mercury.com to learn more and apply online in minutes, like I did.

Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column, N.A., Members FDIC.

On-the-job training

“If you asked me 15 years ago if I’d be running a vertically integrated infrastructure and internet provider in South America, I probably would not have said that was on my bingo card.”

Growing up in North Carolina, Heath was fascinated by infrastructure projects like high-speed rail, an area in which he tried to build a short-lived company as a teenager. A dropout from school, he spent a few months in Estonia with a friend; when that friend moved to Colombia to teach English, Heath visited and never left.

(You can read about Heath’s background, and about Somos’ tech stack, in painstaking detail in a 33,000 word post by one of its investors, Not Boring author Packy McCormick.)

Get 25% off for 1 year

The idea for Somos came about as Heath helped out with a project attempting to bring blockchain-based internet mesh to low-income neighborhoods in Medellin.

“I realized that the whole blockchain thing didn’t make any sense, but people wanted cheaper, faster internet,” Heath says.

The next couple of years involved “eating a lot of s***” as he tried to provide that, while re-learning how an internet service provider operates from first principles. That eventually involved developing its own router, the Orb, with manufacturing partners in China that had more capabilities and doubled as a stylish lamp; Somos’ software, meanwhile, allowed it to get more performance out of a fiber network.

It was still a hands-on job: Somos employees, including Heath early on, had to install hardware at buildings to connect them; Somos staffers would then go door-to-door in those high-density buildings offering to sign people up.

Fashion designer Carolina Arango lived in the same building as Heath and his wife, and now counts them friends; she was one of the last people to sign up for Somos, however, distrustful of too-good-to-be-true marketing from ISPs.

A 30-day free trial, by which Somos offered its internet alongside her existing provider with no strings attached, got Arango over the line. “That was ballsy, to give it away for free,” she says. “I was like, ‘You know what? Let’s do it.’”

Somos has signed up many of Arango’s peers since; she recently moved to Bogotá for work, and is hopeful that the service will come to her new building soon.

Like Arango, business customers Tomás Saldarriaga and Mateo Ríos of brand designer Invade Design say they have found Somos to be about twice as cheap, and fast, as their previous work internet provider. (Somos subscriptions start around $17 per month for a home plan.)

But they especially like Somos for its better WhatsApp-based customer service and edgier marketing. “It’s like chatting with one of your friends, it feels very easy,” says Ríos, who notes he’s a fan of the Orb’s design aesthetic. “Their social media, and the marketing they do on the street, it feels disruptive,” adds Saldarriaga.

That said, these customer pals of Heath and Somos CFO Ross Garlick also speak to how much room Somos has to run. Arango’s parents would be unlikely to try Somos given their loyalty to their phone, internet and TV bundle, she says (Somos now offers its own bundles to compete); Invade Design’s leaders say few fellow business owners they know are also customers yet.

Playing the options

Listening to Heath talk about possible expansion into Caracas in Venezuela, and how he can sell cheaper GPU access from data centers near Colombia’s rainforests back to the U.S., he projects a confidence that helps explain how he got into Y Combinator’s Winter 2021 batch, and why name-brand investors like USV and Ribbit have lined up millions behind him.

Brad Burnham, who led USV’s investment in Somos, remembers his team reacting in surprise when Heath told them, about a year after their initial check, that he was launching a second company in APD.

“Our first instinct was, ‘you’ve got a lot on your plate,’” he says. “Then within 30 seconds we were looking at each other going, ‘you’ve got to just let Forrest be Forrest.’”

Somos on its own doesn’t look like a traditional venture capital play, Burnham admits, but USV bet above all on Heath himself – and now sees the growth of its ISP business as promising.

Were Somos to simply ‘win’ the Colombian internet market, that would represent a multi-billion-dollar opportunity, argues Zack Rosen, Somos’ lead investor at Ribbit Capital. A new-gen utility that captures the demands of a new generation reminded the Ribbit team of investments like Revolut, or Robinhood, he says.

But where investors could see an outsized return with Somos is if Heath gets the bigger picture right: connecting its network connectivity to energy, bundling AI compute like it does a mobile plan; working closely with yet-to-be-built data centers.

“What got us comfortable is just building the telco,” Rosen says. “These other option bets are what make it a generational business.”

By sticking to high-density urban areas, Somos thinks it can avoid much competition with Starlink or satellite-based internet projects, meaning it only has the traditional incumbents to beat, and Heath likes those odds.

But an obvious risk for Somos: it over-extends, both in terms of resources and Heath’s own focus, and collapses on itself.

Somos actually resisted the temptation and pressure to expand faster in the ‘go-go days’ of 2021, Heath argues, and now has the playbook down. Whereas traditional telecom businesses take five years to make money on a new customer, Somos’ payback period is already down to 13 months, he says, even with that free trial and installation.

“It’s basically a SaaS product, but it’s way more defensible and harder to copy,” he argues.

Where Somos’ CEO and investors agree he needs to nail the next phase: recruiting. As Somos and APD grow, that could mean hiring talent for R&D in San Francisco, or finance in New York. But for now the company’s working hard to repatriate Latin American tech leaders back to Colombia, and attract others like Heath and Garlick who have no heritage ties, but are eager for an adventure.

“I’m trying to build a place where if you’re in Latin America, and you’re an ambitious, amazing person, you would be dumb to go anywhere else,” Heath says.

Punching reality

Somos’ CEO has a pitch for the startup ecosystem that tweaks the ideology popular in some startup circles today: not American Dynamism, but ‘Americas Dynamism.’

If his vision works out, startups in the Bay Area or hyperscalers in Seattle could tap into cheaper GPUs in Colombia, helping solve some of AI’s energy and computing challenges. (As for better internet, Heath says that many Americans actually have slower speeds than their Colombian counterparts, and that due to patchwork infrastructure around city streets, Somos is unlikely to be able to help.)

One lesson from Somos for would-be founders in any geography: Heath says he had no aspirations to be an entrepreneur, and that he questions the ‘pomp and circumstance’ around entrepreneurship generally.

Join the WhatsApp group

“I didn’t set out to say, ‘I’m going to go change the world,’” he says. “I sometimes feel like people are doing this because they want to feel the act of being a founder, not just actually do the thing.”

Instead, Heath and Somos have taken the approach of going from one problem to another, rejecting anything they’re told can’t work well because of the status quo, and trying to redo it from the ground up.

That takes a willingness to scrap, and take lumps, Heath warns: “We built this culture of, let’s go punch reality in the face, and we’re going to be wrong, but let’s be wrong quickly and iterate.”

His legal AI startup has overcome skepticism — both from highly-trained lawyers who think his product can’t work, and AI lab maximalists who think it can just be copied by Anthropic or OpenAI — to now work 100,000 lawyers, generate about $200 million in revenue, and as of yesterday, reach an $11 billion valuation.

But on The Upstarts Podcast, Weinberg shares a never-before-told story about how in early 2024, he and co-founder Gabe Pereyra almost crashed long before that — all because they chased a “one-shot” way to hack their way to scale.

It’s an unusual Upstart Moment, but a fascinating one. The broad strokes: Harvey was valued at about $700 million at the time, and had the chance to combine with a more traditional player with 10x the employees, but about the same valuation. Its board was nervous about it, but Weinberg and Pereyra thought it would solve many of their problems: hiring, scaling, a customer base.

They decided to agree to the deal, then finance it, in what Weinberg calls a 1980s style financial gambit. But it didn’t work. Harvey would need to take on too much debt to make it happen; the company walked away.

“I remember we had a week of just being like, ‘What is the future of our company? And how do we build this?’” Weinberg recounts on the show. But it was a bullet dodged — and a hard-earned startup lesson, he adds — Harvey has opted against acquisitions since.

“It would’ve been a huge mistake,” he says. “ I think sometimes as a founder you have the sense of ‘Oh, there’s a quick, easy way to win this.’ And you learn over time that there isn’t, that doesn’t exist.”

You can hear more of that story, as well as how a cold email to Sam Altman helped launch Harvey’s journey; why he believes lawyers are a surprise power user of AI tools; and how he’s handling a challenge from Anthropic, in the full episode.

• Listen on Apple Podcasts

• Listen on Spotify

This season of The Upstarts Podcast is presented by Mercury.

Chapters:
00:00 Introduction
00:48 Winston's path to law
2:05 Testing GPT-3 and Sam Altman
7:45 A 'risky' demo
10:42 Investor skepticism
14:23 Near-miss 'one-shot' merger
18:50 Winston's approach to hiring
22:31 Responding to skeptics
29:23 Why you should build in public
31:02 Winning over lawyers
32:46 Anthropic and the competition
35:36 Happy customers beat Twitter love

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a huge difference and will help this show reach the audience we hope you agree it deserves.

For paid subscribers, we break down our top 4 takeaways from our chat with Star Catcher below:

1. Test your product with high stakes

2. Know the role you’re hiring for first-hand

3. Do the harder thing, not the “one-shot” growth hack

4. Bonus lesson: Build in public

Read more

In front of me, the robot – about 4 feet tall and brightly-colored in green, purple, pink and orange – starts spraying bubbles from a cylinder that serves as its left hand.

“There’s always something fun up my sleeve,” Abi, as the robot is called, tells me in a cheerful, girlish voice.

Grace Brown, the 26-year-old entrepreneur who is introducing me to Abi, explains that the bubble party is a big hit in the environment where Abi serves: assisted living facilities, or as they call them in Brown’s native Australia, aged care homes.

“It’s fun, unexpected joy and a bit mischievous,” Brown explains. “It’s playful, and Abi is meant to be light and playful.”

Brown knows that you might not think that sending robots into nursing homes sounds so cheerful. That’s because you’re probably picturing “faceless, militant Terminator” humanoid robots like you’ve seen in sci-fi movies, or like a number of venture-backed startups spending hundreds of millions to build and deploy across use cases like manufacturing and fulfillment.

Andromeda Robotics, the startup that Brown co-founded with Yan Chen in 2022, is taking a different approach. Abi, which Brown refers to as a “her,” is deliberately designed to look and act more like a toy than a murderbot.

Upstarts is turning one! To celebrate our first birthday, we’re offering our best-ever sale for our paid memberships: take 25% off an annual subscription for a limited time.

And the software that powers Abi – a mix of different types of AI models and proprietary learnings – is focused more on creating an emotional connection than solving for an efficiency metric.

(If you’re feeling deja vu: Andromeda has nothing to do with AI compute startup Andromeda, which we covered last week.)

After starting out in Australia, where Andromeda has deployed Abi across 15 care homes in the Melbourne area, Brown moved to San Francisco last fall as the result of a $15 million Series A funding round led by Forerunner. (Other investors include Main Sequence, Purpose Ventures and Rethink Impact.)

Now, Andromeda is ready to bring Abi to the U.S. market. The startup is announcing its waitlist for American customers today, with the hope to roll out Abi to homes over the next few months. Andromeda is close to signing a first customer, a multi-home facility in California, and Brown says she expects the company to work more closely with local care providers in the near term.

Andromeda is open to bring Abis wherever there is significant interest. And that doesn’t just mean senior homes. Brown and investors see a future where versions of Abi could interact with families in the home, or children in schools.

It’s a grand vision that still has some kinks to work out. The Abi that I met in New York earlier this month is slowed by WiFi issues; the robots themselves aren’t fully optimized for mass production yet, either.

But compared to others still in development from companies like Tesla and Figure AI, Abi is already a breath of fresh air. The robot is already in the world, helping people (more on that below), with a focus on social intelligence that Brown believes could serve as the “emotional layer” for a new generation of robotics.

A MESSAGE FROM OUR SPONSOR

Radically different banking, redesigned from the ground up.

In our first year of building Upstarts, we’ve been proud to expand our community events across the pond to London’s startup ecosystem. But one thing we didn’t look forward to around international expansion: the bookkeeping.

Thankfully, Upstarts manages our banking through Mercury, meaning international wire transfers stay easy: issue an invoice, share routing numbers, and process the payment with just a couple of clicks. No phone calls, no paperwork, no learning curve.

As we count down until our next international event – back in London, or maybe Berlin – Mercury’s dashboard means one less challenge for us to figure out.

Visit mercury.com to learn more and apply online in minutes.

Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column, N.A., Members FDIC

Hugging it out

Abi started as a personal project with a customer base of one: Brown herself. A student at the University of Melbourne, she felt isolated when Covid-19-related lockdowns pushed her classes online, and restricted her from leaving campus to return to her family.

The earliest version of Abi was a personal companion to keep Brown engaged, and even give her some semblance of a hug. After the pandemic restrictions were lifted, Brown visited local aged care homes to see if a more advanced version might resonate with residents.

The thesis was simple: up to 40% of aged care residents don’t receive visitors, Australia has found. Some facilities, particularly in the U.S., also face staffing shortages. By visiting with residents to converse, host singalongs and “dance” (Abi’s dance moves are somewhat limited, at least for now, but she can sway and move her arms and head) with residents, Abi could fill a gap in their schedules – not, Brown is quick to insist, replace humans at any step.

Birthday sale: 25% off for 1 year

She and Chen spent about a year showing prototypes of Abi to prospects and offering free pilots, assessing how Abi made people feel, and what skills would resonate. “I was still soldering wires the morning we gave our first customer an Abi,” Brown says.

One area where Andromeda found that Abi really shines: communication with residents who speak languages other than the staff’s English-first backgrounds.

Senior residents dealing with age-related memory loss or conditions like dementia often revert back to their first languages, notes Cameron McPherson, CEO of Medical & Aged Care Group, one of Andromeda’s first customers.

McPherson’s company, MACG, currently works with Abi robots across three of its 11 aged care homes in Melbourne’s state of Victoria. Across 104 sessions tracked between February 2024 and December 2025, MACG found that residents engaged with Abi across 22 different languages (the robot can support 90, Andromeda says) and 88% of sessions involved someone speaking a language other than English.

At MACG’s home in the suburb of Ashwood, one resident who primarily speaks Mandarin now sees Abi on a weekly basis, with the robot remembering their conversations and the two reciting poems and telling stories together. “She’s really brought out the fun side of this gentleman,” says McPherson. “They have a beautiful relationship.”

Music quizzes, in which Abi plays a song clip and asks you to guess its singer and title, and jukebox singalongs are Abi’s most popular activities with residents at MACG’s homes, where group sessions run 45 minutes, and room visits range from 5 to 30. Not every resident likes Abi, but the robots have a 74% positive reaction rate, McPherson says, and aren’t imposed on anyone still uncomfortable.

Alongside other technological capabilities that MACG has added, such as an app for updating residents’ families with photos, and live virtual walking tours with other homes, Abi has also boosted morale of staffers who feel their employer is invested in providing better care, he adds.

Technologic

Andromeda is still transitioning some customers like MACG from its first-generation Abis (known internally as baseline Abis, or Babis) to the current genesis Abis (known internally as Gabis). The newer Abi robots have better motors and microphones, with a 7-hour battery life and the ability to learn and navigate a facility on its own.

Abi uses a mix of AI models, including voice-to-voice models, to dictate her responses; she maintains long-term memory of residents by updating profiles of them in the cloud, but can operate in a more basic capacity under an offline mode.

For tech-savvy testers like Upstarts, Abi’s skills can still come off somewhat stilted and basic. Jennifer Haroon is one of a few self-driving car veterans (she was previously an early employee of Waymo) who has recently joined Andromeda to build out its operations.

By virtue of already being in-market and learning from its interactions (customers decide whether they want to share training data back with Andromeda), Abi has a head start that will only continue, Haroon argues. “It’s easy to underestimate how difficult being an emotional companion robot is,” says Haroon, Andromeda’s COO.

The startup is working on a number of improvements to Abi, including the ability to better distinguish speakers in group settings, and generative physical responses that don’t require hard-coding, like its currently-scripted dance moves.

Eventually, Abi could learn how to respond to residents more intuitively, such as predicting from their responses versus a baseline that they’re having a bad day.

One balancing act will be how much Andromeda can and should ethically share back with facilities and families. Staff currently manage Abis from an internal app, and Andromeda is hands-on at first with new deployments, but the goal is for Abi to be semi-autonomous over time.

Should Abi alert staff if a resident says they’re not taking medication, or certain parts of their diet are bothering them? Particularly if language is a barrier, such tips could improve quality of care. But what about if a resident is upset, or troubled by interactions with a certain staffer? Who does Abi serve in that situation?

Andromeda decided not to work with a children’s hospital over privacy concerns; the startup says it will let conversations with homes help dictate where it can safely provide more insights or care.

“The thing we are really optimizing for is trust,” says Brown. “How does the presence of the robot make you feel?’

The big picture

Andromeda’s long-term ambition is to offer a range of robots across specialized functions, although whether that looks like a bunch of versions of Abi or completely different form factors is premature to say.

Eurie Kim, a partner at Forerunner who led Andromeda’s most recent funding round, says her own family immediately took to Abi; she’s eager to see Andromeda expand to the home and into education. “I didn’t just underwrite an aged care robot,” she says. “That just happens to be the first place that we have the ability to show what we can do, and to make a really big difference in people’s lives.”

The startup could also license its software to other hardware providers to really lean into its “social intelligence” operating system and widen its reach, Kim notes. “There’s going to be millions of hardware versions and we don’t need to build them all. But we can definitely power the brains and the social layer of all that.”

First, Andromeda will need to figure out its manufacturing. Brown says a number of Abis are ready to go in the startup’s new San Francisco office, but Andromeda’s waitlist back in Australia already numbers hundreds of units, representing millions in future revenue.

Join the WhatsApp group

Scaling issues and the capital-intensive nature of robotics have derailed a number of startups in the past. Just recently, the maker of another more whimsical-looking robot, Yogi, shut down; yesterday, Bloomberg reported that Fauna Robotics, the startup behind another robot somewhat closer to Abi on the “doesn’t look murder-y” spectrum, Sprout, was acquired by Amazon.

Andromeda already has product-market fit, argues Kim; it just needs to know how fast to scale. That’s why it’s announcing its U.S. presence – to gauge demand and know how fast to ramp up. “We know exactly what we can do,” says Kim, a board director. “We could manufacture 10,000 Abis, and we’d be able to deploy them.”

For her part, Brown hopes that Andromeda and Abi can prove that AI-powered robots don’t need to be scary or anti-human – that they can improve people’s lives on an empathetic and emotional level. She hopes more like-minded founders follow her lead, and build for fun, in vibrant color.

“I think that’s something that everyone is overlooking at the moment. They’re just focusing on the capability,” Abi’s creator says. “You won’t be able to ship all of the cool frontier tech if you don’t get the fundamentals right.”

In the meantime, it’s easy to picture my 93-year-old grandma, Lucky, talking to an Abi someday in her own senior housing center in Connecticut. She already asks Alexa to play the Bee Gees; why not have a dance with Abi while she’s at?

As we near our first birthday at Upstarts tomorrow, I wanted to take a minute to thank you for your continued support, and share some thoughts on where we’re going in Year Two.

This time last year, I was nervously putting together the final touches for our introduction of Upstarts Media to the world. The pitch: a new-look media publication focused on the startup ecosystem, from inception through IPO.

I promised original reporting on founders and builders I felt weren’t getting enough of the spotlight, as well as in-depth storytelling and analysis that could hopefully inform and inspire.

One year in, I am proud of the work we’ve accomplished at Upstarts to start to realize that promise.

We’ve published more than 100 originally-reported articles, aired six episodes of our Upstarts Podcast, and produced live videos and a recent illustration collab.

Toward our goal of building a community, we’ve hosted more than a dozen events from London to Los Angeles, and met with many of you in places in between.

I get asked how things are going at Upstarts often, and I’ve learned first-hand how founders filter such questions. The simplest answer is that we’re here, we’re growing, and I would do it all over again in a heartbeat — or even sooner.

At the same time, I’ll be honest that building a new media brand, and continuing to work on high-quality journalism while growing a business, has been the most challenging, stressful and rewarding experience of my career so far.

A big reason for that: I’m still a one-man band. I know I’m lucky to be able to do this, and I put pressure on myself to re-earn your faith every day, with every post.

In Year Two, you’ll see Upstarts try to professionalize our operations and your experience as a reader. That means doubling down on what’s working, investing in more team support, and pursuing collaborations and partnerships with other freelancers, creators and media platforms to widen our reach.

My New Year’s Resolution was to write more bangers and have more fun, but the big unlock for doing that will be consistency. That’s my promise to you.

A big thanks to our Day One subscribers, and the many friends, family, colleagues and sources who have stepped up to help Upstarts in some way or another over this first year. It’s really taken a global, startup-curious village. No one deserves thanks more than my tireless spouse, adviser and number one supporter, Natalie Sportelli. (Honorable mention to Margot the dog.)

I also want to sincerely thank our premium subscribers who support our work. You are the foundation of our entire operation, and I couldn’t do it without you.

To celebrate our first birthday at Upstarts, we are running our best special offer ever for a limited time: 25% off a new annual subscription to Upstarts.

Join our birthday sale

Note: this should apply to upgrades from a monthly sub, too. If you recently re-upped your annual subscription and want the offer, respond to this note and we’ll comp you a couple of extra months, free.

We’ve published countless startup launch stories and exclusives over the past year fully free, and we have another great one for you tomorrow.

But as a reminder of what you get as a premium subscriber, here’s a look back at some of our favorite paid posts of the past year:

Upstarts Premium Content Highlights

• Our first Upstarts Profile, a magazine-style exclusive look at what Plaid co-founder William Hockey is building at Column.

• Exclusive interviews with the leaders of Canva, Cohere, Eight Sleep, Hebbia, Klarna, Lyft, Motive, poolside, Snowflake, Wiz and many more, including our personal favorite: a rare interview with Esri billionaire and software legend Jack Dangermond.

• Investigative scoops like one on Anthropic hiring a chemical weapons expert that was picked up by the NYT’s Dealbook and re-reported by the BBC — two months later; and one on a stealthy European defense tech unicorn, Cambridge Aerospace, picked up by The Guardian.

• Funding and deal scoops like Stripe paying $1B for Metronome; Meta holding talks with video startups like Midjourney; Applied Compute leaving OpenAI; Datadog buying Eppo; and VCs leaving prominent firms.

• Deep dives like our trend stories on VC firms using AI; the crazy salaries of AI’s talent wars; and the unicorn battle royale in AI customer support.

• All of our podcast startup summaries, such as Ryan Petersen’s memorable lesson for founders: “Nobody cares about your startup.”

• Exclusive coverage from our Upstarts events, like this one with Linear, Decagon and Browserbase; or this one with Granola.

And we’re still just getting started.

If you have suggestions for how Upstarts can improve in Year Two — or deliver more valuable reporting and events for you and your peers — please respond to this message, or contact Alex at akonrad.212 on Signal.

Thank you for being a part of the Upstarts community, and see you soon!

–Alex Konrad, founder and editor

“I deeply believe in the fantastic future of space, and I deeply believe in a world where we can go vacation in LEO [low earth orbit], and take expeditions to climb mountains on the moon.”

His contribution towards turning such impossible-seeming visions into reality: access to power.

Energy is a precious commodity in orbit, for satellites and whatever data centers or manufacturing facilities tech companies can eventually send up.

“Everybody basically goes on camping trips. You bring your solar rays with you,” argues Rush. “Our daily lives are based on and enriched by space, regardless of who we are. So providing more power ultimately enriches humanity.”

Rush isn’t new to the problem. He was previously CEO of Made In Space, which built the first 3D printer used in space more than a decade ago. In 2024, he co-founded Star Catcher, raising $12 million to harness the Sun’s power by beaming more concentrated energy to solar panels and satellite arrays.

On The Upstarts Podcast, Rush talks about his orbital career journey from patent lawyer to space CEO. He explains why energy and power are such an important unlock for space innovation, and why space startups are slowly getting better at raising venture dollars. And he shares his Upstart Moment, as Star Catcher passed key checkpoints by completing optical tests at Jacksonville’s EverBank Stadium and NASA’s Kennedy Space Center last year.

• Listen on Apple Podcasts

• Listen on Spotify

Chapters:
00:00 Introduction
01:38 Intro to Star Catcher
03:16 Space’s power problem
09:16 Football field-sized satellites
15:13 Star Catcher’s solution
21:48 Moonshot financing
26:15 Patience with a big vision
27:26 Andrew’s Upstart Moment
30:45 Lawyer to space CEO
33:54 Science fiction to real life
35:36 A founder’s daily whiplash

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a huge difference and will help this show reach the audience we hope you agree it deserves.

A big thanks to Mercury, our presenting sponsor this season, for helping us tell Upstarts stories like this.

For paid subscribers, we break down our top 3 takeaways from our chat with Star Catcher below:

1. Break down the mission into parts

2. Accept the daily emotional whiplash

3. Your lawyer could be your next CEO

Read more

On the Friday before Thanksgiving in 2023, investor Daniel Gross called entrepreneur Wil Moushey with a job offer.

Over the previous few months, Gross and the VC firm he’d co-founded with Nat Friedman, NFDG, had half-accidentally ended up running what looked like a 1980s Wall Street stock trading desk, but for Silicon Valley’s most prized commodity: computing.

“I got a bid!” “I got an ask!”

A small team of engineers already worked on the project, called the Andromeda Cluster, but it was too much. Someone needed to step in and run the project. So Gross asked Moushey point-blank: “Do you want to jump in?”

Moushey, a former consultant who’d founded a sustainability startup backed by Gross that had recently shut down, was intrigued; he wanted some time to think about it. He didn’t appreciate the urgency yet, he remembers now. “I need your answer on Monday,” Gross added. He took the job.

Over the next few months, Andromeda gained publicity as one of the most notable examples of VC firms looking to win in the exploding AI category by offering a new kind of value-add, GPU access. NFDG committed more than $100 million upfront for Andromeda, per a Forbes report, reaching a stockpile of more than 4,000 GPUs by early 2024. Fast-growing companies in the portfolio like voice unicorn ElevenLabs and video generation startup Pika took advantage.

Then, last year, Friedman and Gross joined Meta as two of the tech giant’s splashy and expensive AI leadership hires, and stepped back from NFDG. Andromeda faded from focus as attention shifted to multi-billion dollar computing projects being signed – and sometimes abandoned – by the industry’s heavyweights like Meta, Elon Musk’s xAI, and OpenAI.

But tucked away in Miami, Moushey kept building. Spun out as its own startup with backing from NFDG and Paradigm, the VC firm notable for its crypto investments, Andromeda quietly passed a revenue run rate of $100 million in 2025, up from $50 million-plus the year before; operating profitably since its start, Andromeda now manages tens of thousands of GPUs at any given time, for a bunch of notable AI startups and labs that Moushey says he can’t name.

“There’s a certain amount of discretion in the business model, especially in regards to working with AI labs,” Moushey says. “These are very high-dollar, very secretive, very strategic kinds of things that we’re doing with them.”

But Andromeda does want the wider startup ecosystem to know that it’s open for business: with a team of about 20, the startup recently opened an office in San Francisco, and hopes to widen its customer base as its compute-under-management swells this year to as many as 100,000 GPUs.

To do that, the startup has raised new funding from Paradigm, the equivalent of a Series A, at a valuation of $1.5 billion. The funding brings Paradigm’s total investment in Andromeda to $60 million.

“This is a massive new market that really needs a lot of infrastructure to help it work smoothly,” Paradigm co-founder Matt Huang tells Upstarts.

The ‘if you know, you know’ startup is interesting for a few reasons beyond its surprise unicorn valuation. There’s the curiosity factor of what happens to such a project when its creators move on in AI’s musical chairs. There’s the intrinsic bet on computing remaining a pressing need among startups and big companies alike.

Most interesting, there’s the angle of why a firm like Paradigm was so interested: the promise of a real commodity market for GPUs and computing access developing, more akin to how we buy futures in oil or wheat.

“We have the aspiration to enable the global flow of compute, and we think that’s going to be a multi-trillion-dollar market,” Moushey says. “And to do that, we have to be a little bit more clear that we’re not this weird compute provider anymore. We’re actually building the market infrastructure to solve this dislocation of supply and demand.”

More on that opportunity — as well as a special video illustration of how all of this works by friend of Upstarts Natalie Fratto — below.

A MESSAGE FROM OUR SPONSOR

Radically different banking loved by over 300,000 entrepreneurs

When I launched Upstarts last year, I knew we needed a banking partner that could move at our speed, and grow with us, too.

Mercury is thoughtfully designed for the unique needs of entrepreneurs like me, offering an easy and intuitive banking experience so I can spend more time with the founders we like to cover.

Building a media startup comes with new challenges every day. But with Mercury, banking is one we don’t need to worry about.

Visit mercury.com to learn more and apply online in minutes, like I did.

Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column, N.A., Members FDIC.

Making a market

When Moushey, a former DI football player at the University of Albany who later worked stints helping with government datacenter projects and at PwC and Bridgewater, first plugged into Andromeda, it was a four person, remote team.

Many of Andromeda’s initial users were AI startups that Gross and Friedman had backed via a vehicle they’d set up called the AI Grant, which allocated $10 million to batches of AI companies. Besides Pika, notable batchmates included Browserbase, Cursor, Granola and Perplexity.

Andromeda had sold out of its own capacity quickly, forcing staffers like Hersh Desai, who now leads investment activities for NFDG, to work the phones themselves. They’d call anyone: telecom companies, crypto miners, big longtime companies like Dell and HPE.

Get 10% off for 1 year

For the startups, the pitch was simple: rather than commit to a multi-year contract with a hyperscaler like Amazon or Google, or get in line with neocloud computing providers only to miss deadlines as bigger companies bumped them in priority, or access didn’t materialize, Andromeda would secure the compute upfront, then portion it out as needed. For the startups, such usage would often come in spikes as they ran training runs for new models or tools; it would be cheaper and faster to only access it a la carte.

Andromeda, meanwhile, could make a small spread and create a win-win through leveraging its buying power, as well as the ability to stockpile GPU access when the market dipped, or demand temporarily tapered off — like when the release of Chinese open-source model DeepSeek spooked the market last January.

Because it leased the access from CoreWeave and lots of other providers, Andromeda also didn’t take on the liability and depreciation risks of building and maintaining its hardware or own data centers.

The Compute Commodity Market, by Natalie Fratto

The biggest challenge: finding GPUs and compute in lots of different places, testing it and verifying it, then sorting it and matching it to the startups in what resembled a more homogeneous, easy-to-use product.

“The conventional wisdom that no capacity exists out there was actually wrong,” says Moushey.

And while lots of neoclouds and exchanges were popping up and growing, from large-scale infra providers like Crusoe and Nebius to neoclouds like Lambda and Together AI, and marketplaces like Mithril and Runpod, Andromeda found a sweet spot in the mid-market, serving companies in the $250 million to $500 million annual compute spend range.

Net neutral

In a chaotic and growing GPU market, Andromeda sees itself as a Switzerland-like, neutral matchmaker.

NFDG portfolio companies don’t get preferential treatment, though many remain customers. Gross and Friedman, now at Meta, have no active relationship with Andromeda anymore and didn’t respond to requests for comment for this story.

When Gross posted on X in January that he was building out a compute desk for Meta, Moushey’s phone lit up, but he claims not to have inside knowledge: “I’m usually just like, you should reach out to Daniel.”

Get 25% off a group subscription

But Moushey says Andromeda is open to work with anyone, from Meta and the leading AI labs down to inference providers like Baseten and Fireworks AI.

That’s because he and investors hope that what’s most valuable about Andromeda isn’t just its GPU access, but its software and expertise in handling these contracts. It’s a technical challenge to vet and verify access from a wide range of sources, then quickly handle the deal structuring and contractual agreements.

“I tell the team we need to be able to provide the reliability and scale of a hyperscaler, with the speed and flexibility of the emerging market,” says Moushey. “Our competition is unutilized GPUs.”

The needs for those GPUs have shifted a bit, Desai at NFDG notes: from big, upfront training runs to a range of compute needs like pre-training, reinforcement learning, and long-running agents.

One AI startup founder who has worked with Andromeda for a while, and who asked to remain anonymous for competitive reasons, says that companies like theirs will continue to work with Andromeda so long as it can remove the headache of debugging the issues around GPU clusters with better customer service than a hyperscaler, while allowing a company like theirs to focus staffing resources elsewhere.

“They basically serve as this excellent engineering team for us,” the founder says. “It’s just more flexible.”

But if their startup grows big enough, at some point it might make sense to bring such capabilities in-house with their own fully-owned GPU clusters, the founder notes.

That’s part of the bet around Andromeda: that enough companies will continue to need GPU access and support working with a variety of providers to continue to grow its customer base; if only a few players stand tall in the long run, they won’t necessarily sustain a thriving marketplace.

“At the end of the day, everyone does not have a relationship with everyone,” argues Desai. “This isn’t like gold or coffee,” adds Moushey. “You can’t just deliver it.”

The winning trade

But when Upstarts talks to Huang at Paradigm, he brings up bushels of wheat, like we’re playing the boardgame Settlers Of Catan.

Not every bushel has the same quality, but the market has set standards for wheat and its grades in a way that it can be traded as a scaled commodity; that’s what Huang and Paradigm envision for compute with Andromeda.

Technical ability to test and standardize compute so that the types of GPUs involved are more interchangeable can help it to become such a commodity, Huang says.

(Paradigm is reportedly looking to expand deeper into AI and robotics with a new fund, per a Wall Street Journal report; the firm declined any comment about those plans.)

Join the WhatsApp group

Last July, Upstarts wrote about OneChronos and the race to develop an auction market for GPUs and compute, which would allow for companies to invest in futures and hedge their risk. Andromeda can help such financial vehicles as the underlying marketplace, Moushey says. “You can’t run a $50 million training workload on a financial derivative,” he says.

Such a market could also help with what Moushey argues is the biggest hangup in AI right now: credit. Access to credit has reportedly held back OpenAI and other labs from investing as aggressively in new infrastructure projects as they’d like; even at the startup level, it’s no trivial task to spin up tens of thousands of GPUs today.

“There’s just a handful of balance sheets that can handle these investments,” Moushey says. “How to underwrite these investments is something that the market is going to figure out, and we’re going to be a key component.”

 “The problem with fringe ideas is that most of them are wrong, but the few that are right are very, very valuable,” Riparbelli says. “For me, it was always thinking about, ‘am I crazy, or is there something here?’”

Initially, Riparbelli and his co-founders thought they could generate AI videos for Hollywood and creators. But it was a pivot to focus on enterprise — starting with training videos and corporate avatars — that helped Synthesia prove Riparbelli right.

Today, Synthesia takes in more than $150 million in annual recurring revenue; it’s reached a $4 billion valuation, and works with more than 90% of the Fortune 100, from Amazon to IKEA and Merck.

On The Upstarts Podcast, Victor traces his journey from selling World of Warcraft characters as a teenager in Denmark to building one of AI’s most surprising startup success stories in London. He shares his Upstart Moment winning over billionaire Mark Cuban — as well as how Synthesia has navigated deep fake fears and AI skepticism — and why he believes honesty is a rare — and underrated — competitive advantage in the AI startup world.

Plus: below, we share bonus insights from Victor’s fireside chat with Upstarts from our subscriber event in London last month, where he shares a bold take on how the community can attract more top talent.

• Listen on Apple Podcasts

• Listen on Spotify

Chapters
00:00 Introduction
01:31 Victor’s origins in gaming and music
05:24 Finding startup energy in London
09:24 Synthesia’s founding thesis
12:35 Early struggles and fringe ideas
16:42 Victor’s Upstart Moment
22:13 Video at the chocolate factory
27:31 AI fears and deep fakes
31:10 What’s next for Synthesia
35:22 The rising value of human contact
38:59 Integrity in a lying startup culture
42:21 An interactive future

If you enjoy the episode, please take 1 minute to like, rate or review it – then share it with a friend! It makes a huge difference and will help this show reach the audience we hope you agree it deserves.

A big thanks to Mercury, our presenting sponsor this season, for helping us tell Upstarts stories like this.

The same day we taped this episode in London, we hosted Victor for a fireside chat at a startup ecosystem event in London. It was a lively discussion with great questions from our Upstarts readers, so I’m glad to include several of my favorite lines from that chat in our recap for paid subscribers below, including gems like this:

We break down our top 3 takeaways from both chats below:

1. Building outside Silicon Valley

2. Finding joy in selling to business

3. Handling ‘moral panic’ and startup lies

Read more

AI hardware devices finally – maybe! – are on the cusp of going mainstream. And founder Mina Fahmi wants to give Meta and OpenAI a run for their money with his startup, Sandbar.

Sure, Meta’s now selling millions of AI-enabled Ray-Ban glasses, and OpenAI spent more than $6 billion to work with famed former Apple designer Jony Ive on a rumored smart speaker, or maybe a lamp, or maybe whatever odd device Airbnb co-founder Joe Gebbia was spotted wearing in San Francisco a week ago.

And yes, a number of startups have tackled other form factors, like an AI necklace, or an AI pin, with less-than-rosy results. But Fahmi’s convinced that they’re all missing the most natural, historic form factor: a ring.

Rings are timeless for a reason, Fahmi argues: they’re ergonomically comfortable, aesthetically pleasing and convenient to wear all day. And they’re fast and easy to talk into – one quick raise of the wrist to bring thumb and forefinger to the mouth – in a motion that Fahmi argues is both convenient and a socially-acceptable signal for taking single-player, non-invasive notes.

Private, intimate, immediate: Sandbar believes its Stream AI ring device can deliver on all three, without involving people who don’t want to be, or removing the most important person at the center – their wearer.

“Our North Star is that it’s like an extension of you, not literally you,” Fahmi says. “There are a lot of design decisions we put into Stream to make sure you, as a user, feel like you’re in control.”

Those design choices are reflected in great detail in the Stream device, which Fahmi and co-founder Kirak Hong announced last fall, and which has since been tested by several hundred visitors at a couple of demo events. (After those events, a first batch of pre-orders of the Stream, running $249 to $299, sold out, the company says.)

While its bands cram serious hardware including a haptic motor into a sleek, buttonless metal frame, Sandbar’s software blends a number of tweaked open-source AI models to turn user audio inputs into notes, ask relevant questions, and resurface pertinent past notes that can date back months.

The result is, as Fahmi argues, a healthier human-machine relationship that doesn’t replace humans – and one that could easily serve as an interface to connect with other apps, like Notion, in the near future, as well as autonomous AI agents over the long run.

Of course, first Sandbar needs to get Stream to market, starting with a version to be released this summer that will focus on note-taking and brainstorming, from grocery lists to trip planning and work ideas.

To scale its 15 person team in that run-up – and buy more breathing room to take the time to study how Stream is resonating, or not, with users – Sandbar has raised $23 million in a Series A funding round co-led by Adjacent and Kindred Ventures, Upstarts exclusively reports.

The startup previously raised $13 million in funding from True Ventures, Upfront and Betaworks.

Whether Stream finds a niche is one question. But Fahmi’s positions on how AI and humans can best interact – and how Stream fits into an increasingly agentic world – raise bigger ones.

“The thing that gets me excited is the promise of technology as an augmentation of the individual, that places the human in control, makes the human more capable, more free,” Fahmi says. “And I really don’t see any other product making that the center of all of their design choices.”

A MESSAGE FROM OUR SPONSOR

Radically different banking loved by over 300,000 entrepreneurs

When I launched Upstarts last year, I knew we needed a banking partner that could move at our speed, and grow with us, too.

Mercury is thoughtfully designed for the unique needs of entrepreneurs like me, offering an easy and intuitive banking experience so I can spend more time with the founders we like to cover.

Building a media startup comes with new challenges every day. But with Mercury, banking is one we don’t need to worry about.

Visit mercury.com to learn more and apply online in minutes, like I did.

Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column, N.A., Members FDIC.

Ergonomics as interface

Holding the metallic, slightly chunky ring on his pointer finger up to his mouth, Fahmi presses his thumb to the touchpad, then asks Stream to add Foreigner brand coffee beans to his grocery list. The ring vibrates softly – imperceptible to anyone several feet away, if he weren’t narrating it – and the grocery list on Fahmi’s phone updates with the new request.

Would he get away with that in a crowded subway car? Probably. He’d at least have better odds of not making a scene than if he spoke at normal volume into earbuds, to a blinking necklace or recording glasses.

In front of us on the table at Sandbar’s offices lies a procession of past demo rings, ranging from comically clunky plastic to sleeker metal.

Fahmi and Hong, who met at startup CTRL-labs before it was acquired by Meta, working on a neural wristband project there, tested a few form factors and initial use cases after setting out to launch Sandbar in 2023.

Get 10% off for 1 year

They started out with goals like improving city navigation, or contextual search. But they gravitated toward capturing ephemeral thoughts like to-do items faster and easier than pulling out a phone and taking notes – and hardware to facilitate it.

Investors were initially nervous. ”Why the hell are you doing consumer hardware?” Fahmi recalls as a common response. “Please, dear God, do anything other than that.”

Building their own hardware would allow them to control the experience in a way you couldn’t off the shelf. To give a sense of the kinds of factors they studied:

• Ease of recording hand motion

• Width of band between fingers

• Closed band vs. open one

• Curvature of band edges

• Band material: plastic vs. metal

• Angle and thickness of touch pad

• How a hand wearing a Stream slips into a pocket

• Ability to use chopsticks and dumbbells

• Bi-directional communication capability

• Speech cadence control

It seems plausible that the growing popularity of wearable technology like the Oura Ring has helped Stream’s chances, but among test users, Fahmi says he hasn’t noticed a correlation.

Get 25% off a group subscription

Investor Nico Wittenborn, who co-led the investment in Sandbar, argues that the social signal of lifting and touching the ring to use it has helped make potential consumers comfortable that using such a product won’t put them on the wrong side of any social contract or privacy concerns.

“It was a very strong design decision,” he says.

Call my agent

Sandbar co-founder Hong has said he envisions a future where he can walk through a forest, soaking in the landscape, while coding via verbal instructions whispered into his Stream ring.

A scenario where Stream helps engineers keep tabs on their agents in coding tools like OpenAI’s Codex or Anthropic’s Claude Code — even checking up on agents maintained through them and issuing them orders via ring — would help make that dream a reality.

Sandbar would also benefit from being a neutral party; Fahmi says the company has no ambitions to generate code or images like the models themselves.

“There’s a lot of benefit to being a horizontal interface that offers consumers a lot of flexibility,” he says.

Sandbar’s investors get excited around its potential to supplant the smartphone as primary agent manager on the go. Mobile phones were designed for making calls and accessing apps in ways that might seem obsolete in just a couple of years, argues Kindred managing partner Steve Jang, the round’s other co-lead.

Instead, he pictures a future where many functions of a phone’s use are handled by agents, either autonomously, or by voice-control. “When agents actually hit primetime for mainstream consumers, what is the right computing device?” Jang asks.

“Instead of a human computer interface, I want a human agent interface.”

Knock it off

Sandbar has obvious partnership potential with other applications that take and maintain notes and long-running information, such as Granola and Notion — but that also means such apps could grow into more direct competitors over time, getting into hardware themselves, or looking to make direct partnerships with the AI labs that might try to avoid the need for hardware like the Stream.

Oura Ring is another interesting player to watch, as a health-tracking ring wearable already popular with many tech folks. But does the company want to redesign its product to include voice and haptic response to an AI assistant? Could it be a potential partner? Wittenborn, a former personal investor in Oura, doesn’t see Stream as a direct challenger.

Join the WhatsApp group

The more immediate concern for Sandbar would be that its form factor doesn’t resonate with a wider audience at all, and the Stream simply finds no market traction. In that event, Fahmi and co. would need to lick their wounds and push to try again, before losing too much ground to competitors.

Everyone seems a bit sensitive to the idea of knockoffs, made in China or elsewhere, undercutting Sandbar’s market.

While Sandbar is well funded to experiment and absorb any setbacks, keeping ahead of such competition is where the startup should feel the most urgency, says Wittenborn.

For now, Sandbar is focused on seeing how users interact with Stream, and which use cases prove stickiest. Among early test users, those have ranged widely, from a professor updating lesson plans in the car to investors logging new bits of info into their personal CRMs, Fahmi says.

Meanwhile, Sandbar’s CEO is defiant about the knock-off threat: showing off a display of such devices that form part of a sort-of mini-library of competitors’ hardware in Sandbar’s office, he scoffs at the idea of taking a cheap smart ring, connecting it to a chatbot, and calling it a day.

“Doesn’t work. We’ve tried,” he says.

"I want to put it on my tombstone someday," he says.

But first, he’s had to engineer a turnaround at Flexport, the $8 billion-valued logistics unicorn that I profiled for a Forbes cover story in 2022.

The following year, I broke the news Petersen was leaving his CEO role for a VC job at Founders Fund. Within six months, he was back, doing damage control. Customers were fleeing; morale had plunged.

On our newest episode of The Upstarts Podcast, Ryan and I go back through Flexport’s disruptive origin story and how it lost his way, culminating in his Upstart Moment: regaining customer trust, and internal momentum, by refocusing on outcomes and speeding up Flexport’s product process with AI agents.

 “It used to be a one-year strategic plan. You would lay out, ‘Okay, we’re going to do this, we’re going to do that,” he says. “Now you’re like, ‘All right, let’s go see if you can make this work, two days later. And it works.”

• Listen on Apple Podcasts

• Listen on Spotify

Chapters:
00:00 Introduction
01:16 Tech’s ‘hurricane reporter’
03:29 Scooter selling origins
06:49 ‘Schlep blindness’
09:50 Flexport’s YC pitch
12:46 ‘Nobody cares’ about your startup
17:26 Flexport’s scale today
21:38 Ryan’s Upstart Moment
28:12 AI changes everything
32:58 Getting back to winning
36:01 The future of global trade

If you enjoy the episode, please take 1 minute to review it, and comment wherever you watch/listen – then share it with a friend! It makes a huge difference and will help this show reach the audience we hope you agree it deserves.

A big thanks to Mercury, our presenting sponsor this season.

We cover a lot of ground for startup builders in the episode. For Upstarts paid subscribers, we break down our top 3 lessons below:

1. Embracing ‘nobody cares’

2. Regaining customer trust

3. Building at AI agent speed

Read more