Events at SPC

In SF, we hosted Snowflake CEO Sridhar Ramaswamy for a conversation on leadership in the AI age and where value lives once software commoditizes. Soleio hosted a Design Chat with Joel Lewenstein, who leads design at Anthropic.

In NYC, Demo Night filled the room with builders showcasing work from Luma, ElevenLabs, Sentience, Sandbar, and Flora. SPC India closed out its Curiosity 2.0 Hard-Tech Showcase, where teams demoed working prototypes across physical AI, bio, and deep tech.

Fund Updates

It was a landmark month for SPC companies.

• Baseten raised a $1.5B Series F at a $13B valuation, led by Altimeter, Conviction, and Spark Capital, to power inference for the world’s most advanced applications.

• Longshot Space raised $5M from SPC, bringing total funding to $20M, to build a ground-based kinetic launcher that propels payloads to hypersonic velocities and opens a scalable path to orbit.

• Orb has entered a definitive agreement to be acquired by Adyen for $335M. Orb’s usage-based billing infrastructure will unify billing and payments for Adyen’s enterprise merchants.

• Town announced a $55M Series A led by a16z to build Townie, an AI assistant learns how you work and does the work alongside you.

• Qosmic raised $3.33M led by Accel and Prosus Ventures to build the optical highway to the cosmos, the communication backbone for the next generation of satellites.

Member Updates

Product Launches & Company News

• AXION launched their neural headband, which let Ivan—quadriplegic after a spinal cord injury—order dinner himself using only his eyes and his mind, with no surgery or implant.

• Pilot launched Meridian, an AI accounting platform that runs the entire month-end close so firms can grow without scaling headcount.

Personal Projects, Explorations & Publications

• Rohit wrote an accessible explainer on full-duplex voice AI, models that can speak and listen at the same time.

• Jonathan published breakthrough research on reading emotion from neural data.

• Vaibhav released datasets.bot, a curated list of open robotics datasets spanning 32,000+ hours of robot data.

• Noah open-sourced vibethis, a Chrome extension for vibe-coding the software you’d otherwise pay for.

Coming Up

• The Future of Biosecurity with SPC (NYC) July 8

• Demo Night: Internal Design Tools (NYC) July 14

• -1 to Guardant Health with Helmy Eltoukhy (SF) July 23

Recent Content

We released new conversations with Nato Saichek (Longshot Space) on the 15-kilometer space gun they’re building to fling payloads into orbit, and Rasmus Andersson with Soleio on lessons from Figma, software decay, and creating the Inter font.

The work continues.

I work with an AI agent that gets better every week. Let's call him Alex, for simplicity. Alex handles coordination, drafts emails, summarizes requests, scans calendars. About once a week he surprises me with how cleanly he handles something he’s never seen before.

But he’s not perfect. Sometimes he needs the same reminder twice. Sometimes he replies when he shouldn’t, or stays quiet when he should speak up. Three steps forward, one step back.

At the end of every week I run the same calculation. Should Alex get more responsibility?

Every “yes” expands what Alex can do for me. Every “yes” also expands what Alex can do to me.

This isn’t a new problem. Managers and parents run a version of it every day: how much responsibility is the right amount, and how much is too much, too fast? The variable underneath the decision is always the same: Trust.

I spent years at Dropbox, where the first company value was “be worthy of trust.” We repeated it because the math was brutal: a single breach is a company-ending event. Trust takes years to earn and a moment to vaporize. And the way we’ve been taught to build it is starting to break.

Trust Underlies Every Transaction

Step back from AI for a second. Almost nothing happens without trust. Money is a trust instrument. So are contracts, reviews, credentials, handshakes, marriage, and employment. We built every one of them because trust is what lets two parties act when neither has complete information about the other.

Trust is the binding constraint on adoption. A product can be better, cheaper, and more elegant than its alternatives and still lose because people don’t trust it. When something is brand new it has almost no trust to draw on, and earning it has always been slow and expensive.

In the AI era, where products change weekly, wield enormous power, and act more autonomously and more opaquely than anything before them, earning trust is the central problem.

The Trust Equation

The cleanest framework I’ve found comes from Maister, Green, and Galford in The Trusted Advisor (2000):

• Credibility: Do you know what you’re talking about?

• Reliability: Do you actually follow through?

• Intimacy: Do you promote emotional safety and security?

• Self Orientation: Are you in this for yourself, or for me?

This equation is elegant and has aged well… for humans.

Because that’s who it was built for: lawyers, consultants, doctors, bankers. People you choose carefully, work with for years, and judge slowly. In that world trust accumulates, expensively, over time.

But the equation is missing something. It measures how much trust a person supplies. It says nothing about how much the situation demands. Trusting a teenager to change your tires isn’t the same as trusting a surgeon to operate on your child. Both require trust, but the bar is wildly different.

So let’s stretch the equation. Think of trust as a ratio between the trust supplied (by a person, a product, or an agent) and the trust demanded by the action in front of you. You act when supply clears demand: T_S > T_D.

Supplied Trust (T_S) is built upon:

• Credibility (C): Brand, credentials, reputation, track record. The accumulated signal that you know what you’re doing.

• Evidence (E): Outputs you can experience directly. Demos, free trials, sandboxes.

• Alignment (A): The sense that the other party’s incentives are compatible with yours.

Demanded Trust (T_D) is built upon:

• Stakes (K): How bad is it if this goes wrong, and how hard is it to undo?

• Opacity (O): Can I see inside the system while it runs, and reconstruct why it did what it did afterward?

So action, the actual decision to proceed or adopt, scales with the ratio:

The original equation maps almost perfectly onto the top of this one. Credibility stays. Reliability becomes Evidence. Intimacy is a flavor of Alignment, and Self-Orientation is just its inverse.

What the original missed is the bottom: trust isn’t only supplied by an actor, but demanded by a circumstance.

The Evolution of Trust In The AI Era

Put a pre-AI decision (hiring a consultant) next to an AI-era one (handing an agent your inbox), and every term in the ratio moves.

Credibility doesn’t suddenly stop mattering but it matters less. The signal has gotten noisier. Fake Amazon reviews have been a problem for a decade. While that predates mainstream AI, AI pours fuel on this issue. When you can drop Tom Cruise into a movie he was never in and have it look real, “I saw it with my own eyes” stops being proof.

Evidence is cheap and instant, but never comprehensive. An agent can do the right thing ninety-seven times and, on the ninety-eighth, confidently send the wrong email to your most important customer.

Alignment is the headline question in the era of AI. The canonical worry is the paperclip maximizer. Tell an agent to build the most efficient paperclip factory and, taken literally, it might start a war to clear the way. While it may be doing exactly what you asked for on the surface, you were actually never aligned on the things that matter. At SPC, we’re now explicit enough about this that our agents name a human as their “manager.”

Opacity thickened. Traditional software is deterministic, the same input causes the same output, and when it breaks you can step through it line by line. AI doesn’t work that way. When Alex tells me he’s learned from a mistake and won’t repeat it, I can’t set a breakpoint. I can only ask him how he knows, and then decide whether to believe his answer.

In the AI world the trust variables are fuzzier, the error bars wider, and the components of trust are themselves harder to... trust.

So, Do I Let Alex Run My Inbox?

Back to the question I started with.

The numerator is actually fine. Alex is capable, and I have plenty of evidence of it. The models behind him are credible. He seems well aligned, treating me like his manager.

It’s the denominator that stops me. Email is high-stakes: a misfired message can do real brand damage, especially since the person on the other end doesn’t know Alex is an agent. Alex works around the clock and I can’t watch around the clock. While he’s good at replaying what he did, he’s still murky on the why.

So if you’re building AI tooling and bootstrapping trust, the usual advice is to increase the trust you supply:

• Credibility: Borrow trust from partners, customer logos, and real testimonials until you’ve earned your own.

• Evidence: Build low-stakes, accurate sandboxes. Be radically transparent about your methodology and your prompts. Show, don’t just tell.

• Alignment: Tell me exactly what you’ll do with my data, and how my success becomes yours. An agent saying “yes” is not the same as an agent that wants what I want.

But the real leverage right now is in the denominator, and far more people need to pull on it. Decrease the trust your users have to spend:

• Stakes: Build undo into the workflow. Let me edit after sending. Give me a trial. Walk me up a ladder of graduated responsibility instead of asking for the keys on day one. Be crystal clear about read versus write access at every layer.

• Opacity: Make the logs easy to pull. Cite sources. Allow instant replay. State your assumptions. Share the spec.

Every point you shave off the denominator is a point of trust your user no longer has to manufacture before they say yes.

Interested in SPC? Apply to join us here.

Great ideas aren’t found in charted territory. Reaching them means wandering off the well-worn path and staying lost long enough to find a new frontier.

That doesn’t mean going alone.

Curiosity compounds, and it compounds faster in good company. The people you explore alongside widen your surface area for discovery. Their problems send you somewhere you wouldn’t have looked. The more you ask, the more you find.

South Park Commons exists to push the frontier by gathering the right people. The questions we explore at SPC are indicators of where the frontier ends up next.

If you’re thinking about any of these questions, reach out to the SPC team member below, or apply to join SPC directly.

Can AI express more than just intelligence?

AI is on the trajectory to be the world’s most intelligent lawyer, doctor, and accountant. But empathy, curiosity, connection and humanity are integral to those professions.

• Is it possible for AI tools to truly develop empathy in a way that would be deeply felt by humans interacting with them?

• How can AIs foster a deep connection with humans?

• What is the form factor that AIs should have in order to enable connectivity?

Reach out to Aditya

What markets emerge when supply is no longer constrained?

Many successful companies emerged by unlocking previously inaccessible supply through making it discoverable, accessible, and economically viable. Expertise can now be replicated, services can be partially automated, and a single person can serve dramatically more customers than before. Markets that were previously constrained by scarce human expertise may suddenly become viable.

• What marketplaces were historically impossible because supply was too scarce?

• Which marketplace ideas failed historically because they arrived before AI made the economics work?

• Which marketplaces become more defensible as AI makes software easier to build?

Reach out to Danh

What needs to get built for a world that stays abundant as AI turns energy into a scarce, contested input?

In 2005, Thomas Friedman called it a flat world: cheap energy, free trade, stable geopolitics, abundant labor. Recently, our world has started to look more jagged than flat. One example: AI training and inference are turning electricity into a contested input just as stationary battery costs fall along a Wright’s Law trajectory, pushing home storage toward a default appliance.

• What new tools, regulatory changes, or business models are needed to get breakthroughs to market faster?

• What are the components, materials, models, and software that underpin critical infrastructure?

• If enough homes localize storage, does the grid’s fixed cost collapse onto fewer customers until the system unravels, and who profits from re-bundling the defectors?

Reach out to Evan

What does it mean to learn and how should AI reshape pedagogy?

New education models promise a hyper-personalized and accelerated pace of learning. As AI makes information more accessible, instilling depth of understanding isn’t guaranteed.

• How much of comprehension depends on effort, on the friction of figuring something out yourself?

• Is memorization an outdated proxy for understanding, or is it a foundation we’re too quick to dismiss?

• How do you build AI tools that guide and inspire curiosity without short-circuiting the process?

Reach out to Arian

How do we help every enterprise run its own fleet of models?

The default assumption is that enterprises will rent intelligence from a handful of frontier labs. That assumption is fragile. Open-source models are less than a year behind frontier models, post-training is getting cheaper, and, most importantly, companies are finally figuring out how to capture organizational context and put it to work.

• How can companies turn proprietary context into proprietary models? What verticals benefit the most from this shift?

• What does the infrastructure stack (routing, evals, versioning, observability) look like for running hundreds of specialized models in production?

• What parts of this get built in-house versus outsourced—what new roles emerge inside enterprises, and who serves the long tail of companies without ML teams?

Reach out to Finn

Where are the robots?

Despite a huge influx of capital, we’re nowhere near widespread deployment of robotics in complex, generalized environments. ‘World model’ has become a punchline in funding circles for its imprecision and overuse. It seems plausible we’ll get recursively self-improving AI before we get a robot that can load a dishwasher.

• Are world models necessary for widespread deployment of general robotics, and if so, what remains unsolved to make them work?

• What still needs solving in hardware and the supply chains even when the software layer works?

• What kinds of infrastructure changes (in software, hardware, and the built environment) will be needed to safely deploy more capable robots?

Reach out to Jonathan

What is the future of proprietary intelligence?

Nearly every consumer app is built on roughly the same intelligence, available to everyone at falling prices. But the data that feeds that intelligence will remain proprietary, and problems surround temporal memory: fast decay, inaccuracy, and volatile costs.

• What will be the right architecture for long-horizon user memory?

• If users own their personal memories, how will they be stored, ported and retrieved?

• What does this mean for privacy and regulations—how will data-portability rules, deletion rights, and user-owned-memory standards evolve?

Reach out to Prateek

What does an AI-driven shift toward consumer-controlled medicine look like?

For decades, tech-forward patients and advocates have tried to wrest control of their health choices and data from the medical establishment. AI may radically expand the ease of accessing medical data and its value, while collapsing the cost of reasoning.

• Will we see new AI-driven biomarkers used to continuously manage biological systems?

• When AI medical advice is good enough and nearly free, do consumers start bypassing doctors entirely for routine health decisions?

• Can we build data consent infrastructure as simple as open-source licensing that makes mass participation frictionless?

Reach out to Mark

Are smart devices finally smart enough?

Smart devices promised to rewire how we live, and instead were clunky interfaces with marginal utility. With models small enough to run on-device, hardware can perceive, reason, and act locally.

• Who owns the platform layer when everything is ambient?

• If on-device AI removes the cloud dependency, does that unlock new monetization or does it just make a commoditized hardware product slightly better?

• Does the first mover define the protocol and data layer for everything that follows, or does each category fragment into its own silo?

Reach out to Dylan

In a world where work becomes optional, where does meaning come from?

Work can provide structure, status, belonging, and motivation. Universal basic income can replace a paycheck and none of the rest. The hard problem isn’t money, but meaning. Assuming the struggle that accompanies a universal basic income, there will be a market for the provisioning of meaning itself.

• What provisions break when work, as we know it today, goes away?

• Can you create artificial arenas to provide meaning, like many simulation games already do?

• What is a secular, buildable version of religion?

Reach out to Rohan

Interested in SPC? Apply to join us here.

We’re excited to announce Ryan Atkins has joined the SPC team as Partner, Operations. Ryan’s background spans across dimensions (literally, he spent the first decade of his career teaching physics).

Ryan also spent 10+ years scaling engineering & product teams at Asana, Stripe, and Dropbox. He was the founding site lead for Asana’s Warsaw office, growing it to over 150 employees.

We spoke to Ryan about his love for operating systems and culture building and where he’s most excited to contribute at SPC.

What does the Operating Partner role look like at SPC?

Ryan: My primary role is focused on improving the systems and programs that allow the SPC community to thrive. My goal is to help the partners help members, and help members help each other. This means everything from how we source and market candidates, new member onboarding, to helping members make sense of their own journeys. Day to day, this means anything from moving the furniture to re-writing SPC’s values. Nothing is too small or too big.

What does your past experience look like?

Ryan: I grew up in the Bay Area, studied physics at Stanford, and then did a master’s in education. I figured I’d teach for a year or two, and I ended up doing it for ten, at Eastside College Prep. Teaching taught me things I still use every day—how to build organizational systems, and how to get the best out of a group of people. What I loved was the depth. You go deep with students, you get to know what makes them tick and how to help them grow. You’re limited by breadth though, because a classroom is small. That contrast is what eventually pulled me toward software, because of its ability to reach so many more people.

The transition happened through Aditya [SPC Partner], who I’ve known for about 20 years through college friends who’d ended up at Facebook. I told him I was burnt out from teaching, and he invited me to interview at Dropbox where he’d taken the CTO job a year earlier. I joined and helped build and scale the cultural systems there, straddling both the people and engineering organizations, somewhere between an operational generalist and a technical program manager.

From there I went to Stripe, which was around 1,000 people when I joined and growing fast. I spent about a year and a half in a similar role, sitting between engineering strategy and the people-team cultural investments, on what was broadly called the Education Team.

Then in 2020 I moved to Asana. I was looking for a smaller organization and more engineering-directed work. I came in to lead engineering operations and served as Chief of Staff to the CTO, and ended up staying for nearly six years. I helped scale the org, moved into an Engineering Director role overseeing the adoption and growth of the product, and eventually relocated to Poland as the founding site lead to launch our Warsaw office which we grew to about 150 people.

What drew you to SPC?

Ryan: I’ve been following the SPC team for a long time, and I actually visited back in 2017. So a lot of this was familiar to me. There’s this gravitational force behind the talent density here, it’s hard not to pay attention to that. What I love—whether it’s software or the design of our own team—is operating systems. SPC is a genuinely unique version of that: a small team paired with a large community. It’s the best of both worlds. SPC is a small team that builds at lightning speed, but also has real problems of scale. On top of that, we’re at a crazy moment in technology. The boom in AI is creating seismic waves in entrepreneurial opportunity. It’s very exciting.

What’s something most people get wrong about operations and scaling?

Ryan: People too often pit operations against strategy, like you’re either an executor who does the work, or you’re the one strategizing about it. But operations has a strategy of its own. There’s a strategy to how you operate. The choices and decisions you make reflect your values and goals. You can be highly strategic in an operational role. The other thing is that culture is an incredibly important variable. You can scale in fairly predictable patterns, but so much of what underpins an operating model comes down to culture. Stripe, Asana, and Dropbox all felt very different. What I respect and value is the way a founder’s DNA percolates through a company with their personality—how they act, behave, and what they believe.

What’s a lesson that shows up over and over, regardless of company?

Ryan: Every organization tends toward disorder as it evolves. Entropy is real. For example, there’s an explosion of noise that every organization eventually has to reckon with. Part of the job of an excellent operator is to increase the signal. Sometimes that means amplifying the signal, sometimes it means decreasing the noise. The best operators do both. And repetition doesn’t spoil the prayer. You just have to say things over and over, finding novel ways to do it. That’s true for a teacher in a classroom and it’s true for a CEO in a company.

You spent the first decade of your career teaching physics. What from the classroom still shows up in how you work with builders today?

Ryan: Ten years of trying to get 13-year-olds to listen to you was survival of the fittest. The evolutionary pressure of a classroom forces you to learn every trick for keeping people engaged. You learn to listen and to meet people where they are. The culture becomes the infrastructural layer you build everything else on top of. You need the foundation. The goal is to build structures that make it easiest for our founders to follow the best path, make it hard to screw up and easy to succeed through tools, communication patterns, programming.

Across all of my roles, I’ve always worked on onboarding. For a teacher, the most important day, by far, is day one. What standards do you set? What rituals do you establish? On my first day, I’d hand each student a textbook along with a seating chart, and I’d take the time to pre-write every student’s name—adding them to the legacy of those who held the book in prior years, helping them feel like a part of history.

A year from now, what would make you feel like your work with the SPC community really mattered?

Ryan: I think about my impact the way we think about life-changing technology and how it makes you ask the question, how did we ever operate and build without this? You can’t quite remember or imagine the time before. Success looks like everyone operating at a higher level of execution than before, and feeling good about the work they’re doing.

It’s a lofty goal and hard to pin down, but ultimately the purpose of this organization is for the companies and people we invest in to be thriving. I hope everyone feels great about SPC. Everyone who works here should feel like this is the best work of their life, and everyone we invest in should feel that SPC inflected their trajectory, accelerated their ability to have impact, changed the lens through which they see the world.

Interested in SPC? Apply to join us here.

Amit Jain, cofounder and CEO of Luma AI & SPC alum, joins South Park Commons Partner Finn Meeks to explain why the most defensible AI companies aren’t just building tools, they’re developing systems that can understand and simulate the world.

Amit traces Luma’s path from a free 3D capture app designed to gather training data, through an early bet that video generation was finally viable, to the unified model architecture powering Luma Agents today. He explains why most AI products still break at execution, why foundation labs blur the line between product and research, and why they are the blueprint for the companies of the future. Watch select clips below.

Will LLMs become a thing of the past? World models are strictly better.

Video generation alone isn’t enough. There’s a reason apps like Sora failed.

“In a foundation lab, there is no product or research.”

Minus One is series about the winding journeys the world’s most interesting people take to becoming great—and what they do when figuring out a question we all face: What’s Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.

We spoke with both Harshit and Rohan about their experiences and what they’re most excited about working on at SPC. Read on!

Harshit Madan has spent more than a decade building inside India’s startup ecosystem. At Meesho, he led product and later operations & supply chain through a period of enormous growth, worked on product at Housing.com, and founded his own company, Alle.

In our conversation, Harshit shares how he reads markets, where he thinks the next wave of Indian companies will come from, and the founders he’s hoping to find.

What drew you to South Park Commons?

Harshit: I’ve spent 10+ years at startups. I was lucky enough to be part of Meesho when it scaled from 300 to 3 million orders a day. And then got the chance to build Alle, starting from scratch. Both these stints helped me experience 0 to 1 and 1 to 100 with a front row seat. After all that, I wanted a way to give back, to share what I’ve learned about finding PMF, scaling orgs, and knowing when to press the pedal versus when to pause. SPC felt like the right place to do that, given how seriously it takes early-stage exploration in the Indian ecosystem.

Where are your largest areas of expertise?

Harshit: My background spans consumer tech, product, operations, and supply chain, mostly from the inside of high-growth startups. At Meesho, I led product for two years and then Ops and supply chain for the next two, which gave me a full-stack view of what it takes to scale. At Alle, I got a very different education. What PMF actually looks like, and just as importantly, what it doesn’t. I’ve also tried starting up on my own, which gave me a much deeper appreciation of 0 to 1. I think about org-building, system design, and when to hire aggressively, versus when to slow down and figure things out first.

What’s your thesis on where the interesting opportunities are?

Harshit: I look at opportunities from two angles. The first is market change: when something structural shifts and unlocks a new wave of behavior. Meesho’s early growth was a good example: Jio connectivity created a massive reseller market in tier 2 and 3 cities over a short period. The second angle, which I’m even more excited about, is technology unlocking experiences that simply weren’t possible before. WisprFlow couldn’t have existed without great speech-to-text models. These are magical, never-before-possible experiences. The ongoing AI revolution is creating more of them every day.

Why are you bullish on the Indian ecosystem?

Harshit: India has a high-quality builder pool with deep software expertise, and rapidly growing hardware talent base. Ather and others are creating that muscle. But beyond talent, I think India is at an inflection point where it can produce genuinely global companies. Every country that’s made the leap to developed status, has had a generation of companies that became global giants—BYD and ByteDance from China, Samsung and Hyundai from Korea. India needs that generation, and I think the conditions are there: the talent, the demographics, and market segments that are genuinely representative of global demand. I’m most excited about Indian builders who are aiming at the world, not just the local market.

What kinds of founders do you most want to work with?

Harshit: I want to work with founders who are mission-driven, customer-first, and genuinely obsessed with the problem they're solving. I also think a lot about "why now?" — so I'm especially drawn to founders operating at inflection points, whether that's a shift in the market or in what technology suddenly makes possible. After a decade of learning the hard way, I'd love to share what I've picked up across as many founder journeys as I can.

Rohan Choudhary is a builder and product leader who has spent 14 years building across ad-tech, consumer-tech, and SaaS, with a focus on products built from India for the world. At InMobi, he helped build the core performance ad-tech stack behind one of India’s largest global technology companies. He later co-created Glance, a consumer product distributed through deep OEM integrations, which became a unicorn in its own right and scaled to 400M+ active users globally.

Rohan shares how he sees building in India changing and the trends he’s most excited about.

What drew you to South Park Commons?

Rohan: Years after my MBA, I switched between four jobs in 1.5 years, (including a startup attempt) because I was trying to find work I’d actually enjoy. What I kept coming back to was this – the same 24 hours can go toward building something genuinely ambitious and aligned with who you are, rather than just chasing the next available market opportunity. SPC is one of the few places in the startup ecosystem that takes that seriously—the “where to play” question, well before the “how.”

That question matters even more now. Building has gotten so much easier that there’s a real pull toward “I have a hammer, let me find a nail” rather than doing a more thoughtful search. I wanted to be around people resisting that pull.

Where are your largest areas of expertise?

Rohan: A few things I keep coming back to. The first is taking products global. I’ve done this multiple times at InMobi and Glance. There’s something I genuinely love about navigating the tension between what makes a product work in India, and what it needs to succeed everywhere else.

The second is products that become ecosystems, multi-sided networks and marketplaces. The journey from a useful tool to something with real network gravity is where I feel most at home. Over the last year, I’ve been working on AI commerce ideas built around consumer, designer, and artisan ecosystems, and I have some views on how these get built differently in an AI-native world.

Third, and this one goes back a long way - I almost did a PhD after engineering. I spent a lot of time in ML research at IIT Delhi. At both InMobi and Glance, I really enjoyed working closely with data science and ML teams. I think I’m good at helping researchers identify the right gaps to go after and then figuring out how to get that to market with an actual business model.

And more generally, I have a habit of thinking a few steps ahead—where the puck is going—and using that to make product strategy bets that pay off. At Glance, we were constantly evolving our PMF based on how we thought the market and consumer behavior would shift. That kind of sequencing is something I enjoy.

What’s the current state of the startup ecosystem in India?

Rohan: The first wave of Indian tech was operationally led. The technology was largely a commodity. The real moat was in aggregating fragmented supply, building distribution, and deploying enough capital to make things convenient at scale. India was the market being served.

What’s shifting now is where the value lives. The moat is moving from operations into the technology itself. Indian founders are building frontier IP in AI infrastructure, physical AI, health tech, manufacturing, and not just localizing someone else’s product. The shift I find most exciting is India going from being the market to being the builder.

What investment trends are you excited about?

Rohan: Deep tech finally feels like a breakout category in India, backed by real capital and real ambition. Space and aerospace is another one. Some very young founders are attempting genuinely audacious things here. A wave of IPOs is also creating a new generation of entrepreneurs and angels who’ll keep the energy going.

But the thing I’m most excited about is health tech via AI. People are now going after large, long-unsolved problems—the kind that can meaningfully change quality of life and longevity at scale. That feels different from what came before.

What kinds of founders do you want to work with?

Rohan: Three things, honestly. Agency and resilience—the next few years will keep disrupting propositions and business models faster than ever, so the ability to keep hustling through is non-negotiable. Truth-seeking—I work best with founders who are genuinely trying to figure out if what they’re doing is real, even if the answer is uncomfortable. And ambition—this is one of those rare moments when trajectory-changing companies are actually possible, and founder ambition is the main thing limiting that.

What directions of ideation are you drawn to?

Rohan: Network-based consumer products are still underrated to me. As tools get commoditized, products that require humans on both sides to work maintain something more durable. Agents can turbocharge the moat, but the human core is what keeps it real.

I’m also drawn to what I think of as “moving up the averages”—AI genuinely narrows the gap between tier 1 cities and the rest of the country in healthcare, education, governance. Founders working on that are doing something I find meaningful.

And I’m watching the SaaS model with a lot of curiosity. I imagine a world where every company has its own custom software, deployed by what look more like firm-specific developers than traditional SaaS customers—with a master product that evolves based on fragmented customizations downstream. It’s some hybrid of service and product that doesn’t have a clean name yet, and I’m excited to work with people figuring out what sustainable differentiation even means in that world.

Finally—mental health and suffering. This one is personal. I think AI has a real shot at alleviating a lot of daily lived suffering for a lot of people, and I want to spend time on it.

Interested in SPC? Apply to join us here.

Events at SPC

In NYC, Soleio hosted a Design Chat with Dylan Babbs, SPC alum and current CTO at Profound. In Bangalore we hosted -1 to Razorpay, where Harshil Mathur unpacked the challenge of building payments infrastructure at planetary scale, and -1 to Space, with actual astronaut Shubhanshu Shukla.

In SF, Demo Night showcased explorations at the edge of what’s possible: model memory research, tactile hardware-software interfaces, and new operating systems for an agentic world.

The month closed with the SPC Embodied AI Hackathon, where teams designed and completed a challenge to bring intelligence into the physical world involving manipulation, locomotion, simulation, or flight, with bonus points for collaboration. We ran SPC as a fully-stocked robotics lab including humanoids, drones, SO-101 arms, and 3D printers. From the Desktop Robot Assistant, which moved a candy bar using voice agents, VLMs, and ACT policies, to the autonomous drone that tracks your squats and coaches your form in real-time, many teams had end-to-end policies running on real hardware by the end of the weekend.

Member Updates

Product Launches & Company News

• Cognition raised over $1B in a Series D at a $26B valuation led by Lux Capital, General Catalyst, and 8VC.

• Recursive raised $650M at a $4.65B valuation led by GV and Greycroft, building AI that recursively improves itself via open-ended algorithms.

• Vori raised $22M in a Series B to build the self-driving OS for grocery stores, processing over $500M in payments across 55 cities.

• Goodfire launched Silico, a platform for building AI models with the precision of written software.

• VMax published PopuLoRA, research on co-evolving LLM populations for reasoning self-play.

• Asymptote Labs open-sourced Agent Beacon, a tool for AI agent security.

Personal Projects, Explorations & Publications

• Rex released PathOn Robotics, showing a custom navigation stack running on a Unitree Go2.

• Arya published a piece on debugging multi-agent systems, with a follow-up case study on iterative refinement.

• Jake shared a directory of 12,000+ open-source STEP parts for mechanical and electrical engineering, and released text-to-cad for generating 3D models in coding agents.

Coming Up

• Hard-Tech Showcase | Curiosity 2.0 (BLR), May 31st

• Design Chat with Joel Lewenstein (SF), June 8th

• -1 to Snowflake with Sridhar Ramaswamy (SF), June 22nd

Recent Content

We released new conversations with Nikesh Arora (Palo Alto Networks), Amit Jain (Luma AI), and Steve Ruiz (tldraw) covering everything from why the most defensible AI companies are building world models, to the infinite canvas as a use case for agents, to the cybersecurity crisis ahead.

Thanks for reading! The work continues.

The next six months in cybersecurity will be unlike anything the industry has faced before. Nikesh Arora, CEO of Palo Alto Networks, joins South Park Commons partner Aditya Agarwal to discuss why AI is both the threat and the only viable solution.

Nikesh traces his own unlikely journey from Google's Chief Business Officer to leading one of the world's most critical cybersecurity companies with no prior cybersecurity experience, shares what Masayoshi Son taught him about risk, and explains why most founders are too de-risked before they ever start building. He also breaks down why the same AI models writing code today will soon expose decades of hidden vulnerabilities at a scale no IT team is prepared to handle. Watch select clips of the episode below.

The cybersecurity crisis is here. Nikesh explains why AI is both the threat and the only viable solution.

Every great tech company starts with product. Most fail when they stop prioritizing it.

Everyone had the tech behind ChatGPT, but no one thought it was a real product.

Minus One is series about the winding journeys the world’s most interesting people take to becoming great—and what they do when figuring out a question we all face: What’s Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.

Nobody in Silicon Valley cares what you did a decade ago. Even more so now, when a month in AI time can feel like years in SaaS time.

This ecosystem runs on obsession with the future. It doesn’t matter if you started a company twenty years ago and took it public. If you’ve stopped building, you’re forgotten. There’s an egalitarianism here which I haven’t found anywhere else in the world. That quality is Silicon Valley’s superpower.

But lately, I’ve watched it curdle into something destructive.

The average tenure at startups keeps compressing. I see brilliant researchers cycling between labs every eighteen months, never staying long enough to push past the frontier they wanted to explore. Founders wind down companies after a year, not because they tested the idea and lost conviction, but because the initial excitement wore off and something shinier appeared. I see engineers treat career progression like a video game. Collect the title, vest the cliff, move to the next stage.

To be clear, there’s good speed and there’s bad speed.

Good speed is table stakes. It’s strategic velocity, when you take fast signals seriously and pivot from stagnant ideas. Learning cycles get compressed and execution compounds, getting you from insight to iteration before the world has caught up.

Bad speed is impatience that pretends to be ambition. It optimizes for novelty over depth and doesn’t let you sit long enough inside the unglamorous stretch where the real edge is formed. Because the truth is that the frontier rarely yields itself in the first six months.

The danger today is that AI’s pace can make constant reorientation look like progress. Every week brings a new model, a new lab, a new interface, a new consensus. Reorientation at good speed can be strategic. But when everyone is perpetually restarting at bad speed, nobody stays long enough to discover something non-obvious. And increasingly, that’s what bad speed looks like: an ecosystem optimized for perpetual beginnings, but allergic to the endurance required for real breakthroughs.

The harder skill now is discernment, knowing when a pivot reflects genuine new information and when it’s simply fueled by fear, impatience, or the seduction of momentum itself.

AI tools make it dangerously easy to generate forward motion. You can prototype faster, test faster, ship faster, and convince yourself faster that you’re on the right path. But lower friction can also mean accelerating false starts and finding yourself in a local maxima.

Finding the right footing at the very beginning matters more than ever. In a world where it’s cheap to build almost anything, the real edge is choosing what is actually worth building and staying with it long enough to learn something the market doesn’t know yet.

Why Depth Always Beats Bad Speed

You don’t really understand a problem until you’ve lived inside it for a while. Trust with a team isn’t built in a quarter. Neither is the kind of institutional knowledge that lets you see second- and third-order consequences before they happen. The most valuable professional assets (judgment, relationships, domain expertise) compound. Cut the timeline short and you never reach the part of the curve where the returns become extraordinary.

When I visited hard-tech companies in El Segundo, where teams are building electric boats, nuclear reactors, and space habitats, what struck me wasn’t just the ambition. It was the time horizon. Conversations there stretch decades, not quarters. Nobody talks about disruption. They talk about what needs to exist and the painstaking work of figuring out how. There’s a seriousness to that kind of building that you lose when you’re optimizing for eighteen-month cycles. It reminded me of old-school Silicon Valley, before everything became a show, when people built things because someone had to build them.

If you only spend a short amount of time on a project, the most you can do is modify an existing solution. You can make something faster, cleaner, cheaper, more polished. But genuine novelty demands a different kind of compounding. You have to compound toward creation, not toward endless modification of what already exists.

In a world where AI is compressing the time it takes to execute, compounding matters more, not less. When powerful tools can handle more of the building, the bottleneck shifts to knowing what to build and why. That kind of judgment doesn’t come from a weekend with a new model. It comes from years of accumulation.

We live in fast times. AI is rewriting the assumptions underneath entire industries. The temptation to optimize for optionality is completely rational. But grit, resilience, and loyalty will be eternal traits of the best people. No amount of speed changes that.

Interested in SPC? Apply to join us here.

This past month SPC members brought hardware to the stage, shipped new products, and sat down with leaders building at the intersection of AI and the physical world.

Events at SPC

We hosted Minus One to Replit, featuring co-founders Amjad Masad and Haya Odeh. Later we were joined by Nikesh Arora, chairman and CEO of Palo Alto Networks, for a conversation on cybersecurity and AI. Nato Saichek, co-founder and CTO of Longshot Space, shared how his team is building a giant pneumatic cannon to shoot satellites into orbit. Amit Jain, co-founder of Luma AI and SPC alum, discussed how he's pushing toward unified general intelligence that can perceive and act in real-world environments.

In NYC, OpenAI, Ramp, Nominal, and Dataland joined us on a panel to break down what forward-deployed engineers really do, and how they're collapsing deployment cycles.

Our recent Demo Night in SF surfaced another wave of wacky and wild explorations, from visual hearing aids to sleepwalking inside generative models. In Bangalore, SPC India hosted escape velocity, bringing together teams building at the frontier of space, defense, and dual-use technology.

Member Updates

Product Launches, Company News & More

• Zain & Eddie launched Flipbook, a prototype that streams every pixel on your screen directly from a model. No HTML, no layout engine, no code.

• Oliver & Ben released Mesa, the first POSIX-compatible filesystem with built-in version control, designed for enterprise AI agents.

• Ben, Grey, & Aditya launched Bevel 3.0, the health platform that unifies your biometrics, medical records, and life context into one AI-powered coach and data analyst.

• Emma launched Scriben, a pen that acts as a second brain for knowledge workers.

• Andrew shared how he turned to AI to research his girlfriend's brain tumor after two surgeries fell short, deciding to push the frontier of medical research himself.

Coming Up

• Human Agency in the Intelligence Age with Jason Crawford (SF), May 4th

• -1 to Snowflake with Sridhar Ramaswamy (SF), May 6th

• Design Chat with Dylan Babbs, CTO at Profound (NYC), May 7th

• -1 to Razorpay with Harshil Mathur (BLR), May 14th

• Embodied AI Hackathon (SF), May 15th-17th

Thanks for reading! The work continues.

We’re entering the next era of AI: post-prompting. As we reach the apex of manual prompting capabilities, the actual frontier lies in autonomous agents and spatial canvases. You should not have to coax an AI step-by-step through a build. You should drop it in a repository, give it a directive, and let it operate in the background. It should push solutions to you, not wait to be pulled.

We spoke with Replit co-founders Amjad Masad and Haya Odeh last month at SPC. The future is clear: we’re entering a new era of autonomy.

1. Infinite coherence unlocks true autonomy. A year ago, agents lost their minds after an hour of operation. Today, with embedded evaluation loops and unit tests, models can stay on track indefinitely.

2. The pre-training plateau forced a pivot. The current leap in agent capabilities is driven by reinforcement learning and models reasoning through their own compressed context windows.

3. Agents must push, not pull. You should not have to coax an AI step-by-step through a build. Autonomous agents must operate in the background, understand your organization, and proactively deliver solutions.

4. The defining bottleneck is amnesia. Today's interfaces force you into a tedious loop—re-explaining your project, re-arguing your intent, micromanaging a tool that forgets everything between sessions. The winning systems will passively retain context, infer intent from your usage patterns, and shift the dynamic from negotiation to expectation.

5. The UI is shifting to the spatial canvas. Typing is too slow. The future of interaction is multi-modal. It looks less like a terminal and more like Iron Man’s lab: a mix of voice, touch, and spatial manipulation.

6. The big labs abandoned UX. OpenAI and Anthropic are building raw engines. They are ignoring the user experience to focus on model scale. This leaves a massive, lucrative space for founders to invent the actual interface.

7. Consolidate or die. Context switching kills velocity. The winning developer environment places the mobile app, the backend, and the investor pitch deck onto a single, unified canvas.

8. Syntax is dead; logic is king. The role of the engineer switches permanently when AI graduates from a Q&A chatbot to a persistent background operator. You start orchestrating complex systems instead of boilerplate.

9. Vibe coding weaponizes domain expertise. The profile of a builder changes when natural interaction replaces rigid prompt engineering. The next great applications will be built by domain experts solving their own complex problems.

Watch the full Minus One talk with Amjad and Haya, out now on YouTube.

Minus One is series about the winding journeys the world’s most interesting people take to becoming great—and what they do when figuring out a question we all face: What’s Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.

Real, working demos are hard. Making them excellent is harder. Our portfolio companies in India stood up to the challenge. Packed room. Epic stories. Energy you can’t find anywhere else.

Here’s a quick highlight reel, followed by the demos from SPC companies.

Maya

You don’t expect a combination of singing, dancing, movie-making and world-building in just one demo. But that’s exactly how Dheemanth Reddy started things off. He and Bharath Kumar are building Maya — the voice interface for the majority of the world to access all frontier AI models.

SurgeGrowth

We all love the idea of scale. But who’s building the structures that will scale in the agentic future? Vaibhav Dusad, Joe Kochitty and Shivam Mishra are—with SurgeGrowth. SurgeGrowth gives operators the command center to design and run complex agentic operations at scale.

GameRamp

Have you ever broken out in hives at the mention of the term “let’s take this offline,” or the dreaded… “circle back?” Thankfully, Shashank Vandgrangi and Vivek Ramachandran from Gameramp are solving for that with Actioneer—the agentic decision layer for growth teams, where AI agents continuously detect opportunities, drive recommendations, and drive action!

Armor

“What do I do with my money? What’s the right decision, from among many possible scenarios?” These are questions that keep us up at night. Thousands of financial products, changing regulations, and uncertain life situations have overwhelmed India’s mass affluent. Yogesh Tomar and Biswanath Patel are building Armor—a proactive financial planner with a hedge-fund grade quant engine, and emotionally aware AI that starts by understanding the person.

SuchamaAI

Manufacturing involves high-stakes environments, where unpredictable changes can lead to massive downstream losses. And yet, factory planning runs on brittle legacy software—hardly built to adapt. Sanjay Shekhawat and Ayush Agarwal are building SuchamaAI—the context-aware intelligence layer for manufacturing that adapts in real time to changing scenarios in man, material, machine, and method.

Fest

Social media promised us connection. It has created anything but. Likes, swipes, endless scrolling... trolling, even. Harish Uthayakumar and Samesh Lakhotia are building for real connection in the age of AI, with Fest — to help everyone feel seen, heard, and understood.

Cent Health

If you or a loved one has ever known what it means to have be dependent on luck rather than certainty for a health outcome—Cent Health is the answer that removes luck from the equation. Shashank ND, Arpit Garg, and Anshul Khandelwal are building an AI-led preventive health intelligence platform focused on early detection of 300+ life threatening diseases—across cancer, cardiac, neuro, and metabolic conditions.

We are excited to be part of each and every one of their journeys, and cannot wait to see what’s next.

Interested in SPC? Apply to join us here.

Events at SPC

March at SPC was packed. We hosted a Minus One conversation with Elad Gil, fireside chats with Stephen Wolfram (Wolfram Research), Thomas Wolf (Hugging Face), and a Design Chat with Sam Stephenson, exploring how Granola approached their recent (and controversial) rebrand. We also hosted Dr. Fei-Fei Li (World Labs), who shared her perspectives on finding a north star and building with intentionality.

SF’s Demo Night showcased explorations from digital public squares to ambient intelligence. Across the globe, SPC India hosted its 2nd-ever Demo Faire in Bangalore—the room was packed for live demos!

Fund Updates

• Sentience raised a $6.5M seed round led by Bain Capital Ventures to build AI that emulates and multiplies individual human minds.

• Nyne raised a $5.3M seed round co-led by Wischoff Ventures and SPC to build people intelligence infrastructure for AI systems.

• Escargot raised a $2.75M seed round co-led by Wischoff Ventures and Hannah Grey VC, with participation from SPC, to build the greeting card company for the 21st century.

We also kicked off our Founder Fellowship S26 cohort. These companies are building across the spectrum. Space. Energy. Robotics. Agent swarms. Quantum. Biotech. Now's the time to be maximally ambitious, together.

Member Updates

Company News & Product Launches

• Luma launched Uni-1, their newest model that both thinks and generates pixels simultaneously.

• Durable launched their AI builder that helps people launch a new business in minutes.

• Simular launched Sai, the always-on AI co-worker.

• Lica opened early access to their platform, allowing you to turn image model visuals into editable layers.

• Noble Machines debuted their autonomous humanoid robot, Moby, at NVDIA GTC.

Personal Projects, Explorations & Publications

• Jason publicly launched Neuron, AI-native messaging.

• JackJack launched PUNK, turning your phone into a remote control for Claude Code.

• Alex’s AI Language won 1st prize in the Scope competition for its real-time VLM approach.

• Lucas published a blog post on modeling physical motion with AI.

• Kion open-sourced CLaaS, a framework for updating LLM weights in real time.

Coming Up

• Building at the Frontier: A Panel at SPC NYC (NYC), March 31st

• -1 to Replit with Amjad & Haya (SF), March 31st

• SPC Demo Night (SF), April 9th

• -1 to Longshot Space with Nato Saichek (SF), April 15th

Recently Released

Evan explores where founders can still build safely as the AI labs expand their reach in Avoiding the Eye of Sauron, and Aditya reflects on what it means when the skill you’ve spent decades mastering becomes free overnight in When Your Life’s Work Becomes Free and Abundant.

Thanks for reading!

Interested in SPC? Apply to join us here.

Most founders are using an outdated map. The SaaS playbook doesn’t work anymore.

We hosted Elad Gil at SPC for a fireside chat a few weeks back. He might be the biggest crusader against conventional wisdom that I know. Here are a few takeaways.

1. Lingering on stagnant ideas is the most common and fatal mistake founders make right now. The market is moving with unnatural velocity. If your product is not gaining immediate traction, the market is giving you a clear signal. Kill it and move on. When a product works in this market, it works quickly. There is no prize for bleeding out slowly over three years on an idea that nobody wants.

2. Culture is winning. Culture is not kombucha on tap. It is not ping-pong tables or offsites. Fast traction is the only culture that matters. When a startup is growing exponentially, every operational sin is forgiven. When growth stalls, every minor friction point becomes a cultural crisis. Focus entirely on moving the metrics. The culture will follow the traction.

3. Foundation models have crystallized into a capital-intensive oligopoly. OpenAI, Anthropic, and Google control the frontier. Open-source models survive primarily through the massive balance sheets of companies like Meta. Competing at this base layer without billions in compute and a hyperscaler partnership is a dead end. The infrastructure map is drawn. Startups must accept this reality and build higher up the stack.

4. Buyers are uniquely open to unproven tools right now. Corporate boards have mandated AI strategies from the top down. Every CEO is asking their team for an AI plan. Big law firms are notoriously slow adopters of technology, yet they adopted Harvey early. This willingness to experiment is unprecedented. The sales cycle has compressed. Startups must capitalize on this temporary window before procurement departments return to their usual friction.

5. Delegation is a trap. Scaling founders are consistently told to hire executives, step back, and delegate. This is a trap. Delegation is necessary for organizational scale, but stepping away from the product is fatal. Builder CEOs consistently outperform. Unapologetic micromanagement is not a flaw. It is a requirement. The transition from maker to manager ruins companies when founders lose their grip on the core product details.

6. You don’t always need a co-founder. Silicon Valley insists every startup needs a co-founder. This is a recent invention. Accelerator programs normalized the equal-equity duo to simplify batch processing. Historical data rejects this entirely. Michael Dell built his company alone. Jeff Bezos built Amazon alone. Larry Ellison drove Oracle by himself. Solo founders can and do win.

7. Incumbent reaction time is compressed. Startups used to enjoy a five-to-seven-year grace period. You could build a wedge before legacy players even noticed you. That window no longer exists. Incumbents are hyper-aware of the AI shift. They are moving faster to deploy capital and clone features. Startups must assume immediate retaliation from incumbents. Speed of execution and deep customer integration are the only ways to survive the compressed reaction time.

8. The single-product startup is dead. Silicon Valley used to preach doing one thing perfectly. That is a luxury founders no longer have. If you build a single tool that is ten times better, an incumbent will build a “good enough” version and bundle it to their existing customers for free. Because AI drastically accelerates how fast your team can ship code, cranking out multiple, deeply integrated products from day one is the only way to survive the incumbent cross-sell.

9. The actual open frontier for startups lies in agentic workflows. Text generation is a solved problem. The immediate technical challenge is moving to systems that maintain persistent memory. These systems must execute complex, multi-step actions across different environments autonomously. That is where new enterprise value will be captured. The winners will not just summarize data. They will execute work.

10. Building software from scratch and fighting for enterprise distribution is no longer the only path. The new -1 phase looks like a buyout. Founders are executing AI-driven rollups. They buy legacy businesses in stagnant verticals, strip out operational bloat, and radically expand margins using AI. It requires a non-consensus triad of skills. You need the ability to buy assets, drive operational change, and deploy deep AI expertise. You buy the distribution, then automate the operations.

Navigating this frontier requires unnatural velocity. The founders who succeed will be those who ignore the outdated map.

Watch the full Minus One talk with Elad on YouTube.

Minus One is series about the winding journeys the world’s most interesting people take to becoming great—and what they do when figuring out a question we all face: What’s Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.

The opportunity has never been greater. Neither has the risk of building the wrong thing. I spoke to the SPC community about the state of the market for founders building in and around AI. Weeks of debate followed, and the conversation continues to evolve.

It increasingly feels like founders are building under the Eye of Sauron. In Lord of the Rings, the Eye of Sauron is an all-seeing force. Once it fixes on you, there’s nowhere to hide. Today, the foundation model labs are starting to feel the same way and the line of sight is only getting bigger.

Competition is no longer one-dimensional. There are now four vectors: incumbents, other startups, the foundation models themselves, and—most importantly—your own customers.

The labs aren’t just building better models. They’re arming your customers.

To understand what that means for you, evaluate your startup across three dimensions.

1. Buyer Capability: Can your customers actually wield the new AI tools?

2. Buyer Agency: Are they empowered to just go build?

3. Product Architecture: Is your product primarily workflow or chassis?

While capability and agency are about your buyer’s skill and autonomy, product architecture asks where your product sits in the stack.

• Workflow lives at the interface layer — the surface users directly interact with to get a job done.

• Chassis is the underlying layer — data, security, orchestration, identity and access, compute, deployment. The enabling infrastructure that everything depends on.

The more your product lives at the workflow level, without deeper roots in chassis, the more exposed you are.

If your buyers are high-agency, high-capability, and your product is mostly workflow, you’re at real risk of them building their own version of your product. I’ve seen it firsthand — internal teams spinning up bespoke tools in days that would have taken months to procure and implement a year ago.

Companies At Risk

Run those dimensions together and a few patterns emerge. This may be sobering, especially if it describes your starting point.

Software for software companies. For founders building horizontal workflow applications for high-agency, high-capability organizations, this is aimed at you. If your customer’s team looks a lot like yours (talented engineers with access to frontier models), you’re in danger. They’ll just build it themselves.

Startups and mid-market tech companies are the most dangerous ICP in 2026. We’ve heard this firsthand from internal teams at startups and mid-market tech companies building their own versions of workflow tools they used to pay for — what would have taken months to procure and implement a year ago is now a weekend project. And it’s not just them. Engineers at multiple frontier labs are doing the same thing.

One conclusion here is that low capability + low agency customers who used to seem “boring” (enterprise-focused, services-heavy, slow procurement) are now relatively attractive. The slog is the moat.

Building around model deficiencies. Look at video and image apps. A year ago, it might have been sufficient to say “it doesn’t do inpainting well” or “character consistency in video is broken.” That kind of thinking is now a losing bet.

Model capabilities are non-linear. The behavior is so emergent that what’s hard today becomes trivial tomorrow.

You should assume that the models will eventually do everything.

Build something that gets stronger as the models improve. Not something they make obsolete with the next release.

How To Survive

Survival hinges on how you respond. It’s no longer whether you can build—it’s whether you know what to build, and why, before the window closes.

Strategy can’t wait. Everyone can build fast now. I met two non-technical founders in New York who built fully-fledged products in two days. That’s incredible, but it also means build speed is table stakes.

The old playbook was to build something cool, raise money, figure out the hard stuff later. You had a few years of runway to figure out your strategy. That’s over. The premium is now strategic speed. Technical founders have to lean into the discomfort of answering questions that used to wait until Year 2. The questions about who you’re really building for and where the puck is going now need answers in months, sometimes weeks. You have to be intentional from Day -1.

Go deep on domain. AI is the ingredient. Domain mastery is the product. Find a space where you understand the customer, the workflow, and the problem better than any lab ever will. Legal, healthcare, energy, agriculture, construction — these are industries delivering services and goods, not building software. They’re low technical capability, low agency, and the enterprise go-to-market is hard enough that the labs won’t bother.

One useful reframe: imagine every lawyer, financial analyst, contractor and nurse as a software engineer with domain knowledge. What would you build that these hybrids would need?

Seek out human problems. Selling. Change management. Trust. Compliance. Network effects. Community. The edge is who can get closest to the customer, navigate procurement, and do the messy work of actually getting customers to discover, buy and adopt technology.

Any place where the problem is fundamentally about people, that’s where to focus.

Build chassis products. Chassis is how you end up on Sauron’s side, not under its threat. There are deeply technical problems (AI research loops, bare-metal deployment at scale, compliance and audit infrastructure, SCIM management) where coding agents don’t meaningfully outperform experts over the next 24 months. At the same time, enterprises want these new capabilities but aren’t built to absorb them. If you can build the systems that let companies safely accept and leverage AI capabilities at scale, that’s a powerful position.

Atoms over bits. AI meets robotics, drones, hard tech, science, and biology — and the rules change. A coding agent can’t weld a joint, fly a survey mission, synthesize a compound, or validate a clinical finding.

The moats here are real: proprietary hardware, physical distribution, regulatory approval, and years of operational data that can’t be scraped. These are categories where the foundation model companies aren’t coming for you…yet.

This isn’t meant to be all doom and gloom. In many ways, this is the most exciting time to build in decades.

Yes, Sauron’s line of sight is larger. But the opportunities have never been greater. Customers are eager to try and buy new tools. Companies are clearing seven figures in revenue per employee. What once required large teams and months of work can now be built by a handful of people in days.

The founders who win won’t just build faster.

They’ll pick problems the Eye can’t see.

Interested in SPC? Apply to join us here.

Grant Lee, cofounder and CEO of Gamma, joins South Park Commons GP Jonathan Brebner to share how the "worst idea ever"—an investor's words before he hung up—became a $2B AI unicorn taking on PowerPoint and Google Slides.

Grant gets into what most founders get wrong when competing against entrenched incumbents, why hiring “painfully slowly” was one of Gamma's biggest advantages, and how to build a team where hard problems feel like play. He also covers how Gamma survived the Silicon Valley Bank collapse with two weeks to go before a make-or-break launch, and why founders who want to build an audience online have to push through “cringe valley” first. Watch select snippets below.

Gamma isn’t focused on competing with other startups, they’re changing how people create presentations in the first place. The real challenge is behavior change.

Early discomfort isn't a sign you're doing something wrong. When you’re sharpening storytelling skills, you have to go through “cringe valley” first. Grant’s audience is now a key growth channel for Gamma.

Three years ago, SVB collapsed two weeks before Gamma’s planned launch. The team earned a few resilience points in the process.

Grant’s Minus One Journey: Gammafied

Minus One is series about the winding journeys the world's most interesting people take to becoming great—and what they do when figuring out a question we all face: What's Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.

Not long ago, I spent a weekend writing code with Claude, Anthropic’s AI assistant. I’ve been programming for more than 20 years. I was one of the first engineers at Facebook, where I built the original search engine. I went on to become chief technology officer of Dropbox, where I scaled the engineering team from 25 people to a thousand. Code has been the foundation of my career, the craft I’ve spent my adult life mastering. And after that weekend, one thing was very clear to me: We will never write code by hand again. Something I was very good at is now free and abundant.

While I was building software with AI, I also noticed that the AI agents elsewhere were building social networks, the very product I helped create at Facebook. Little coding agents were spinning up functional social platforms for themselves. It’s all a bit silly, but what they produce is kind of indistinguishable from what humans built on the larger internet. Both the form and the function of my early career are now being produced by machines.

I sat with that for a while, and what I felt was wonder mixed with a profound sadness. There’s something deeply disorienting about watching the pillars of your professional identity, what you built and how you built it, get reproduced in a weekend by a tool that doesn’t need to eat or sleep.

But here’s the thing about disorientation: It passes. And what replaced my sadness was something I didn’t expect: a kind of wild, almost reckless energy.

In the five days after that weekend, I produced more code than I had in the previous five years. That is not an exaggeration. The software I was making was better than the code I had produced on my own in the past, and far more ambitious. Things I would never have attempted before, because the cost of building them would have been too high, suddenly became possible in an afternoon. I wasn’t watching myself become obsolete. I was watching the constraints I’d accepted my whole career dissolve.

That shift, from sadness to mastery, reveals something important about this moment. The conversation around AI and work has calcified into two camps. The doomers say we’re all going to be replaced, and the boosters say everything will be fine. Neither camp captures what it actually feels like to live through this. The truth is messier. You can hold wonder and grief in the same hand, mourn a version of yourself while sprinting toward a new one.

What I’ve found more interesting than my own experience, though, is what I’ve observed in the people around me. I run South Park Commons, a community and venture capital fund for builders who are figuring out what to work on next. Through SPC, I see hundreds of engineers, founders and technologists navigating this shift in real time. And there’s a pattern that keeps surfacing: The old playbook for evaluating talent is breaking down.

One of our members recently ran about 20 work trials for engineering hires—essentially, extended, weeklong job interviews—and found zero correlation between years of experience and adaptability to AI tools. Another member told me that what predicted success in hiring people who possess that adaptability was evidence of a builder’s disposition: cool personal websites, side projects, an obvious love of making things. FAANG on the résumé and a name-brand university, meanwhile, predicted almost nothing.

A third member shared something even more striking. His company started giving tasks that were intentionally too long to complete by hand during coding interviews. The assignment became a remarkably clean filter: You could quickly tell who was using AI tools in their daily work versus who had merely been reading about them. The gap in the number of lines of code the two groups write wasn’t 10%. It was closer to 10x.

This might sound like a narrow observation about the software industry, but I think it’s something bigger. We are in the middle of what may be the largest shift ever in how knowledge work gets done. And the trait that matters most isn’t intelligence, or credentials or years of experience. It’s someone’s relationship with change—not whether they’ve seen change before, but whether they run toward it.

There’s a common assumption that younger workers will adapt more easily and older ones will resist. But the dividing line isn’t generational—it’s dispositional. Willingness to change seems to operate as an independent variable, cutting across age and seniority in ways that defy easy categorization. I’ve watched 15-year industry veterans pick up these tools and absolutely crush it, while some recent graduates treat AI as an abstraction to be debated rather than a tool to be used.

As an investor, this realization has reshaped what I look for in founders. The people I’m most excited about aren’t the ones with perfect pedigrees. They’re the people who seem constitutionally unable to stop tinkering, who get antsy when things stay the same, who treat every new tool like a puzzle they need to solve before the day is over. I’ve started to think of it as the difference between your résumé and your restlessness. I’d bet on restlessness every time.

Silicon Valley has always been one of the most meritocratic industries, but that has never meant credentials and experience don’t matter here. They just matter less. Now they are going to become even less important.

Paul Ford wrote beautifully in The New York Times about vibe coding’s potential to democratize software, to put the power of building into more hands. I share his optimism. But I’d add that this democratization isn’t just about access to tools. It’s a reordering of what we value in people. We spent decades building a culture that worships credentials and experience. Those things aren’t worthless, but they’re no longer sufficient. The new currency is adaptability, and unlike a Stanford degree, it’s available to everyone.

If anything, this shift is teaching me what it is like to be human again—not human in the romantic, AI-can-never-replace-us sense, but human in the uncomfortable sense, the part where you have to let go of the thing you were in order to become the thing you might be.

That’s always been the hardest part, long before AI. The technology just made it impossible to ignore.

This article was first published in The Information.

Aditya Agarwal is a general partner at South Park Commons. He was previously chief technology officer of Dropbox and an early engineer at Facebook.

Interested in SPC? Apply to join us here.

When Tony Xu cofounded DoorDash 13 years ago, he was rejected by more than 100 investors. Today, it’s a $70B+ behemoth dominating the delivery industry.

Tony joins SPC General Partner Aditya Agarwal to reveal how DoorDash won the delivery war and answer the burning question of whether AI agents pose a threat to his company. He also shares why customer obsession became the company’s guiding principle, the challenges of digitizing the physical world, and how startups today can build competitive advantages in the age of AI. Watch select snippets of the talk below.

DoorDash is currently dominating the delivery industry, but what will happen when new consumer AI agents hit the market?

Perfection is not the goal. Not every experiment becomes the final answer. Teams at DoorDash focus on quickly testing and shipping solutions.

DoorDash struggled to raise capital in its early days. Tony personally remembers being rejected by over 100 investors. Today, the company is valued at over $70B.

Minus One is series about the winding journeys the world's most interesting people take to becoming great—and what they do when figuring out a question we all face: What's Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.

Events at SPC

South Park Commons celebrated our 10th anniversary the best way we know how: live demos. Amidst the science fair-style booths, founders, members, alumni, and friends gathered to discuss building the future (and see it with their own eyes). At its foundation, our community is shaped by curiosity and the willingness to stay with the unknown. What began as ten people gathered around Ruchi’s dining table, has grown into something enduring.

In celebration of a decade, we shared our 10-year anniversary video—a reflection on what it means to keep showing up before the idea exists.

The momentum didn’t stop there. We hosted a Minus One conversation with Tony Xu (DoorDash CEO) and a talk on Consciousness, Intelligence & Spirituality with Swami Sarvapriyananda. Demo Night in SF spotlighted explorations on the frontier. In Bangalore, SPC India hosted Building for Bharat, a demo night showcasing founders building for the Indian market, and a fireside chat with the Anthropic CTO, Rahul Patil.

Fund Updates

It’s been a remarkable start to the year for SPC companies.

• Baseten raised $300M Series E at a $5B valuation, led by IVP and CapitalG to power model inference for app companies at scale.

• Render raised $100M in a Series C extension at a $1.5B valuation led by Georgian to give developers an intuitive path to reliable and scalable cloud infrastructure.

• Profound raised $96M in a Series C at a $1B valuation led by Lightspeed to build AI agents that take marketing from concept to execution and define how brands are represented in an AI-first world.

• Goodfire raised a $150M Series B at a $1.25B valuation led by B Capital, advancing the frontier of AI interpretability for LLMs.

• Koah raised $20.5M in a Series A led by Theory Ventures to power monetization for AI-native applications with privacy-first, LLM-native ad formats.

• Fulcrum raised $25M in combined Seed + Series A funding led by CRV to build AI-first workflow automation for insurance brokerages.

• DealMomentum was acquired by Salesforce. The conversational insights and revenue orchestration platform will be integrated with Agentforce 360 to ground agents in the voice of the customer.

• ProducerAI was acquired by Google Labs to enable artists to create and refine songs through AI-powered collaboration.

Member Updates

Product Launches & Company News

• Nirva debuted at CES 2026 — an AI wearable designed to be worn all day that journals your life, tracks mood, and offers personalized guidance.

• Kenneth & Oliver launched Bedside, taking coordination work off clinicians’ plates, so they can spend more time delivering care.

• Sunain Labs open-sourced Sunain Gameplay, a ~2M-frame dataset of gameplay data for world models and video generation.

Personal Projects, Explorations & Publications

• Raveesh published a deep-dive on continual learning.

• Bryson wrote a blog post on conditioning diffusion models.

• Ed published research on finetuning multiagent systems end-to-end as a novel workaround for catastrophic forgetting in continual learning.

• Vatsal wrote an article on World Models, drawing from philosophy and ML to offer a fresh perspective on the space.

Coming Up

• Fireside Chat with Julie Zhuo (BLR), March 3rd

• Minus One to Elad Gil (SF), March 5th

• Beyond Belief with Nir Eyal (NY), March 11th

• SPC India Demo Faire (BLR), March 17th

• -1 to World Labs w/ Dr. Fei-fei Li (SF), March 18th

Team Updates

We’re excited to share three team promotions!

• Jonathan Brebner is now a General Partner. Formerly SPC’s Chief Storyteller, he helped shape -1 to 0.

• Mark Jacobstein joins full time as a General Partner after 6 months as a Visiting Partner.

• Dylan Itzikowitz is now Partner and Head of SPC NYC, where he’s led the community’s growth

Recent Content

We released new conversations with Andrew Yang, Waseem Daher, and Anurag Goel on the Minus One Podcast. From AI’s broader societal impact to the day-to-day realities of building companies, these are real looks at decisions in motion.

Thanks for reading! The work continues.

Interested in SPC? Apply to join us here.

“We’re heading towards a job-killing tsunami for the average American.”

Andrew Yang saw it coming before almost anyone else. In 2018, he warned that without a plan to handle job displacement, AI would destabilize society. Now, he says the crisis he feared is no longer hypothetical—and that the political violence and social unrest we’re already seeing are just the beginning.

Andrew joins his early mentor, SPC General Partner Mark Jacobstein, and Aditya Agarwal to unpack where we go from here: what UBI might look like in practice, why coders are more at risk than truck drivers, and why his new company, Noble Mobile, will actually pay you to use your phone less.

Yang was an early AI visionary, but he admits he got the sequencing wrong. As a presidential candidate, he warned truck drivers and call center workers. He now says he should have warned management consultants, law school grads, and coders.

Yang’s early career was shaped by a series of failures. All those experiences taught him one thing: rejection isn't fatal.

The political system is “held hostage” by the 11% who vote in party primaries, so candidates go to the extremes. Andrew has a system fix.

Minus One is series about the winding journeys the world's most interesting people take to becoming great—and what they do when figuring out a question we all face: What's Next? Because before you launch at Zero, you have to figure out what to launch at Minus One. Hosted by South Park Commons Partners.

Interested in SPC? Apply to join us here.