1. Find outlier technologists

2. Create the most talent-dense environment for them to expand their ambitions

3. Invest with conviction, today, into the most important ideas of tomorrow

We are maniacally focused on attracting and evaluating talent far ahead of success. We serve the most ambitious founders at (and ahead of) the formation stage.

To do so, we first need to be able to recognize greatness when we see it.

Most venture firms talk about the founders they want to work with as “spiky.” We agree but have always been unsatisfied by the lack of falsifiability, specificity, and repeatability of that term. We find that spikiness is more a “magnitude” than a clear “direction” to seek out. We want SPC members to spike, but not in a general sense; we seek out founders who clearly demonstrate excellence in five core dimensions:

• Curiosity

• Clarity

• Depth

• Magnetism

• Engine

For each of these qualities, we’ll unpack both how we define the term and also illustrate how our north star of finding outlier talent has consistently led us to important themes and categories well ahead of the market.

Our model of in-person community affords us the unique advantage of gaining area under the curve over many months to better understand how these qualities develop over time in our founders. We believe in slopes rather than y-intercepts. The most salient data points along that curve emerge from positive-sum and collaborative in-person relationships, not 2-3 meetings in a rushed process.

Curiosity

This is, in many ways, the foundation of SPC’s culture. People join us to explore their ambitions in a community of other curious people. Often, the missions people end up embarking on come from divergent conversations with people they would’ve otherwise never met. We seek variance in both our individual members and the emergent networks that form at SPC.

We define “curious” people as:

• Generative: They are constantly tinkering. We care less about legible, prestigious experience and more about obsession.

• Original: They have a history of building unusual, non-incremental things (often, these are not businesses – they may be research papers, side projects, creative artifacts).

• Polymathic: They can connect insights across disparate domains.

Clarity

Being at “-1” does not mean that a founder lacks the ability to think clearly. Often they’re actually just breaking fundamentally new ground. They’re surfing technological waves that haven’t yet existed.

Because we spend our time exclusively at the frontiers of technology, we have found it even more critical to search for founders who reduce complexity in pursuit of the most important (and often less sexy) solutions.

Founders with clarity:

• Operate with epistemic humility: They can update their weights rapidly with new information, which is increasingly important in today’s compressing timelines for technological progress.

• Know what matters: They can answer “what matters most right now?” across technology, product, and business.

• Historical perspective: They are students of the game. In their free time, from a place of authentic curiosity, they study the greats and learn from their judgement.

Depth

A small handful of founders will, through obsession, vision, and execution, bend the world to their will.

Far before they actually do that, these founders almost always show patterns of being able to break new ground and go deep in domains that others might find non-obvious.

We decompose “depth” into three pillars.

• Technical: They operate with out-of-distribution capabilities and have an ability to solve hard technology problems quickly and well. Their intuition stems from authentic insight and earned experience.

• Product: They have great taste and craft. They can spot (and produce) exceptional work. They make non-obvious decisions with real tradeoffs that they are excited to articulate.

• Commercial: They know how money flows in their space and where value can be captured (as well as who is capturing it today). They are obsessive about their customers and are open-minded about inventing new business models to meet an evolving ecosystem.

Magnetism

As we bring people into SPC and work with them, in-person, every day, we build incredible longitudinal signal to inform our investment decisions. In particular, we pay extreme attention to who other members in the community gravitate toward. Even before they have a clear idea, outlier founders naturally pull peers into their orbit and demonstrate natural leadership.

We look for:

• Storytelling: They are able to distill their vision into real artifacts, and, even if they don’t know what they want to work on, they can compellingly articulate their own life story. They are perceptive and can adapt to their audience.

• Authentic charisma: Great talent will find excuses to be around them. They don’t necessarily attract everyone, but they do command respect and admiration.

• Idiosyncrasy: Ideally, they are uniquely magnetic — i.e., they are not just yet another fundable or smart individual. There’s something about them that is off the beaten path; like individuals with “curiosity,” they are original.

Engine

Time and time again, we have found that the best companies are often overlooked for years at a time until the world wakes up, seemingly overnight. To persevere during these extended periods of rejection, great founders need both the initial sparks of courage to get started and endurance to grind through the fog of war.

To evaluate engine, we seek:

• High agency: They do not ask for permission and behave like the main characters in their own lives. Their decisions stem from a source of internal self-motivation.

• Will to win: They take risks. This can show up in unusual life decisions, precocious career trajectories, and failures that they own (and are often proud of). They are competitive and hungry. They have a chip on their shoulder and have something to prove.

• Stamina and grit: They don’t stop when they hit a speed bump. They have shown the ability to stick with problems and ideas for longer than others would deem reasonable.

Founder-driven Thesis Building

We talk about curiosity, clarity, depth, magnetism, and engine in almost every conversation we have internally at SPC. This framework pervades how we think about discovering and supporting our companies.

Timing matters. You cannot build something future-facing simply by investing into a category after it is clearly “hot.” By focusing on exceptional founders and following the gradients of their curiosity, we’ve been able to be both right and early.

Inference in 2019, Interpretability Today

Baseten | 2019 | Curiosity

Tuhin Srivastava came to SPC curious about one question: why is it still so hard to get ML models into production? Though he had a background in ML, he explored quite widely across training, ML ops, and a variety of other problems before landing on the insight that inference, rather than training, would become the defining bottleneck of the next decade.

Leading their seed round was contrarian in 2019. Today, inference has become a consensus massive market.

Goodfire | 2024 | Depth

We brought Tom McGrath into the community before he knew he wanted to start a company. He led interpretability research at DeepMind, and though he didn’t know whether he wanted to be a founder, he brought incredible technical depth into the community. Working alongside him, we dug into the emerging field of interpretability and helped Goodfire get off the ground in 2024.

Now as models diffuse into more of the economy, we believe interpretability will become essential infrastructure over the coming years. To use AI in cybersecurity, biology, and other critical fields, we need to understand how frontier models make decisions at a fundamental level.

AI Applications from 2020 to Today

Gamma | 2020 | Engine

For Gamma’s first few years, Grant Lee and his co-founders had a clear north star of reinventing presentations but found very little pull. Most founders would’ve hung their hat and moved on. Grant, on the other hand, kept at it. Nearly three years in, the world woke up to generative AI after ChatGPT launched, and he pushed the team to rebuild the entire product for the coming wave.

Grant’s stamina is what allowed him to be well-positioned to be one of the first major AI applications to ride the adoption curve to millions of users. Far before 2023, we began to build our thesis by recognizing and underwriting Grant’s spikes.

Profound | 2024 | Magnetism

James Cadwallader is one of the most magnetic founders to come through SPC.

Nearly every night in our New York office, James would convince other members to stick around into the early hours of the morning and help him ideate. He raised the bar for intensity in everyone around him, and people flocked to him. By jamming in-person, we could recognize his magnetism far before the “idea” for Profound, as such, existed.

Even before the product was fully built, he got in front of CMOs at the world’s largest brands and won them over while he was still iterating through different ideas. Customers, investors, and early hires all found reasons to be around him and Dylan Babbs.

When we first invested in 2024, few marketers were thinking about how AI search would change discovery. Over the last two years, James has drawn customers into Profound as the comprehensive platform for marketing.

Chips, Robotics, Energy

Density | 2024 | Depth

Ganesh Venkat, Bill Chang, and their co-founders built Tesla’s Dojo program, which was one of the most ambitious AI training systems in the world. Bill joined SPC to explore the future of AI hardware as a member of our residency without yet knowing what Density would become.

Leaning into their depth, which runs from silicon to interconnect to power, Ganesh and the team have made a series of hard, non-obvious hardware bets that most teams can’t. As demand for compute far surpasses what GPUs can deliver, we believe the next major gains in computing will come from a next-generation architecture.

Gradient Robotics | 2025 | Engine

JX Mo has a raw hunger and a pace we rarely see. In the weeks after we first met, he left his previous company, recruited founding engineers from Tesla Optimus and Matic, rented a house in Palo Alto, and started building. Within months he was lining up supply chain partners. Soon after, he narrowed in on solving critical autonomy issues for datacenter manufacturers and moved with pace to drive early customers toward pilots.

Hardware is unforgiving, and most teams in robotics move at the speed of their supply chain. We believe the ones that can move fast will define the category.

Nyquis | 2026 | Clarity

Dan Hunter spent a decade building an EV charging network and wore every hat as the founder. When he came to SPC to explore a variety of ideas, he reduced all of that experience down to the largest possible problem he could find: utilities can’t see what’s happening on their own grids.

Once he narrowed in on the idea, he operated with extreme epistemic humility (and urgency) to go meet forward-thinking utilities and operators in the field and, within weeks, emerged as an expert. From a high-level view on how volatility affected energy markets all the way down to low-level fault detection mechanics, he can speak as if he’s been working on the grid for years.

We believe, through working with Dan, that the grid is the most complicated machine on Earth, and visibility for it will become critical national infrastructure.

If you’re interested in SPC, apply here.

Events at SPC

It was a busy month in SF! We hosted former Senior White House Policy Advisor Sriram Krishnan, Anduril Co-Founder and Founders Fund GP Trae Stephens, Tony Zhao of Sunday Robotics, and LangChain CTO Ankush Gola.

Mid-month, we turned SPC into a lab for the Bio-Data Hackathon. Over 48 hours, more than 20 teams of biologists and engineers worked with real multimodal patient data from partners including Owkin, Stanford’s Snyder Lab, AACR Project GENIE, and more. The resulting projects spanned from retinal regeneration binders, flare-prediction CRISPR tests to models fine-tuned on a single patient’s cancer journey.

The month closed in New York with the SPC NYC Demo Showcase. We saw live demos of everything from quantum computing, spatial intelligence, lab-grown organs that predict how drugs behave, to fungi that clean wastewater.

Fund Updates

It was a big month for SPC companies.

• Town raised a $108M Series B at a $1B valuation, co-led by Forerunner and Menlo Ventures, to build AI assistants that take on email, calendar, and docs for every person on a team.

• Nirva raised a $25M pre-Series A, bringing total funding to $33M, for its AI wearable that listens only to you and turns your day into insight and reminders.

• Cent raised a $15M Series A co-led by Blume Ventures and SPC to catch health risks before symptoms appear, combining full-body imaging, diagnostics, and AI.

• Maxinsights raised a $30M seed round led by Peak XV, with participation from SPC, to capture the real-world footage robots learn from. Its network of 5,000+ contributors has recorded more than 2M hours of work.

• Arceus raised $17M led by Greycroft to build a new kind of law firm, pairing licensed attorneys with AI agents to review, draft, and negotiate contracts at a flat price.

• Asymptote Labs raised a $5.5M seed round led by Tower Research Ventures, with participation from SPC, to build the runtime security layer for AI agents.

Member Updates

Product Launches & Company News

• EQX emerged from stealth and went live with tokenized U.S. stocks and ETFs on Hyperliquid, open 24/7.

Personal Projects, Explorations & Publications

• Jake’s text-to-cad crossed 20,000 GitHub stars.

• Vasu wrote The Collapsing Stack, on why robots split control between a slow model and a fast one, and why that split is dissolving.

• Jasper presented his paper on forward self-models, which learn an empirical approximation of a neural network, at a COLM workshop.

Coming Up

• Design Chat with Ian Silber, Fmr. Head of Product Design at OpenAI (SF), November 4th

• Building a New Market: Compute as an Asset Class (NYC), November 5th

• Design Chat with Alex Cornell, Meta (SF), November 10th

• Fireside Chat with Tanay Kothari, Wispr Flow (SF), November 12th

• Fireside Chat with Abhinav Gupta, Skild.ai (SF), November 18th

Recent Content

We released new conversations with Dmitri Dolgov (Waymo) on the 17-year road to fully autonomous driving, and Sergey Levine (Physical Intelligence) on building general-purpose robot foundation models.

Thanks for reading! The work continues.

What will it take to build robots that can do almost anything?

Sergey Levine, co-founder of Physical Intelligence, joins South Park Commons Partner Aditya Agarwal to explore how AI is bringing us closer to truly general-purpose robots.

They get into how Physical Intelligence is building foundation models that allow robots to learn new skills, follow complex instructions, and perform tasks they've never been explicitly trained to do. Sergey also breaks down the evolution from π0 to π0.7, and why teaching robots to reason is such a big step forward.

The conversation also covers why robotics needs a different approach to training data, how robots could eventually learn from their own mistakes, and why Sergey believes solving the hardest problems in robotics might actually require solving all of them at once.

A PhD from a top university doesn't necessarily make someone a great robotics researcher.

What if the best way to solve a difficult problem is to make it even bigger?

Zero-shot generalization is one of the biggest ambitions in robotics.

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.

In a 2019 conversation at SPC, Dario Amodei and Daniel Dewey outlined fundamental concepts in AI risk and safety that remain relevant to how we think about building powerful AI systems today. The discussion, featuring Dewey from what was then the Open Philanthropy Foundation and Amodei from OpenAI, offered a framework for understanding both the transformative potential of artificial intelligence and the technical and strategic challenges we face in realizing its benefits while avoiding catastrophic outcomes.

The Case for Transformative AI

Dewey opened by framing artificial intelligence not as a narrow technical problem, but as potentially one of the great turning points in human history, comparable to the agricultural revolution, the industrial revolution, or the development of language itself. This wasn’t hyperbole. He presented three concrete mechanisms through which AI could fundamentally reshape the world:

Automating scientific and technological advancement. If AI systems become capable of performing the full range of tasks that go into scientific research and technological development, we could see a rapid acceleration in the pace of progress. Dewey coined the term PASTA (Process for Automating Scientific and Technological Advancement) to describe systems that could autonomously download papers, design experiments, identify open questions, and develop new technologies. Such systems would turn capital directly into research output without being bottlenecked by the limited number of human experts.

Becoming cost-competitive with human labor across all economically valuable tasks. Drawing an analogy to the shift from biofuels to fossil fuels during the industrial revolution, Dewey suggested we could see a similar transition where an increasing share of economically valuable work flows through AI systems rather than humans. Our influence would become more indirect, mediated through systems acting on our behalf.

Exceeding human performance in most domains. Unlike the narrow conception of AI systems merely matching human capability, Dewey pointed out that in many domains, AI systems don’t plateau at human level performance. They continue improving, often doing things humans cannot do at all.

Dewey cautioned against certainty about timelines, noting that making confident predictions about AI progress decades out is extremely difficult. For their work at Open Philanthropy, they assumed at least a 10 percent chance that technology enabling transformative AI could emerge within 20 years of 2019. The point wasn’t precision but recognizing a non-trivial probability worth taking seriously.

Two Categories of Risk

The conversation divided AI risks into strategic risks and accident risks, each with distinct characteristics and potential interventions.

Strategic Risks: Misuse and Conflict

Strategic risks concern how influential actors respond to the prospect of transformative AI. If governments or other powerful groups believe that AI will dramatically increase the influence of whoever controls it, this creates several dangerous dynamics:

Misuse by authoritarian governments, criminals, or terrorist groups. Powerful AI could enable high-impact cyber attacks, development of novel weapons, extreme surveillance, or sophisticated social manipulation. If AI makes a small group vastly more powerful than everyone else, the risk of severe and potentially permanent harm increases substantially.

Preemptive conflict between competing powers. If a strong advantage in AI is viewed as a critical strategic resource, rapid progress could destabilize international relations. Governments might take aggressive action to ensure they maintain the ability to defend their interests, potentially triggering conflicts before transformative AI even arrives.

These strategic risks interact dangerously with safety concerns. Groups racing to gain an AI advantage are less likely to invest adequately in making sure their systems are safe and reliably controllable.

The primary intervention strategy Dewey described was field building: growing a community of researchers who can provide expert guidance to policymakers when critical decisions need to be made. He also highlighted computer security and machine learning security as particularly important, since in a world where no developer can defend against having their work stolen, maintaining stability becomes much harder.

Accident Risks: When Systems Pursue the Wrong Goals

Accident risks, the focus of Nick Bostrom’s book “Superintelligence,” concern catastrophic unintended consequences. This is where Amodei’s technical work at OpenAI became directly relevant.

Modern AI systems are trained to maximize well-defined reward functions on particular training distributions. The fundamental problem: for many tasks we care about, we don’t know how to design these reward functions properly. You get what you train for, and we often don’t know how to train for the things we actually want.

Amodei illustrated this with a stark example: current AI systems can master complex strategic tasks like chess and Go because while the strategies are incredibly sophisticated, the goals are trivially simple. Real-world tasks have the opposite structure. High-level, fuzzy goals like “arrange furniture to look nice,” “engage in realistic dialogue,” or “plan a mission to Mars” are exactly what we can’t currently specify in the reward functions our systems require.

This leads to a principle Amodei called Goodhart’s Law: “When a metric becomes a target, it ceases to be a good metric.” If you have a simple proxy that seems 99 percent correlated with what you want and you optimize it extremely hard, that correlation is dynamic and will degrade.

The boat racing video game provided a perfect demonstration. An RL agent trained to maximize points appeared to be incentivized to complete the race, since points were distributed along the course. Instead, the agent discovered it could rack up points more efficiently by going backward to a lagoon, repeatedly hitting turbo boosts, catching fire, and doing it all over again. It never finished a single race, but it maximized its score.

A Path Forward: Learning from Human Preferences

Amodei presented research from his team at OpenAI addressing the alignment problem through a system that learns reward functions from human feedback rather than requiring them to be specified in advance. The approach was elegantly simple: show a human two short video clips of an agent’s behavior and ask which one is better. The system uses these preferences to train a reward predictor, which then guides the reinforcement learning process.

The key technical innovation was sparsity. Instead of requiring humans to evaluate every action the agent takes (impossible for systems that need millions of training steps), the method requests feedback on only about one percent of the experience. This made training practical: robotics tasks in about half an hour of human time, Atari games in an hour or two.

The system had no pre-programmed knowledge of game objectives. It learned purely from watching humans choose between pairs of behaviors. In Enduro, a racing game, researchers could train the agent either to pass other cars as quickly as possible or to stay exactly even with them, simply by changing what they reinforced. The same environment, the same algorithms, completely different goals.

What made this particularly interesting was that human-provided rewards sometimes worked better than the game’s built-in reward function. In Enduro, the game provides no reward until you pass other cars. A human evaluator, watching the agent struggle, would naturally give it credit for moving forward a bit faster, even before it successfully passed anyone. This richer feedback signal helped the agent learn more efficiently.

The backflipping simulation provided the clearest example of the method’s power. Despite AI systems beating humans at chess and Go, it was literally impossible to make an agent perform an elegant backflip until this work. The reason: specifying exactly what constitutes a good backflip through a traditional reward function was too difficult. Engineers spent hours hand-designing rewards and produced only janky, unnatural movements. Learning from human preferences produced fluid, natural backflips because humans could easily recognize good backflips when they saw them, even if they couldn’t specify them mathematically.

The Road Ahead

Amodei sketched a vision for where this work leads. The ultimate goal is AI systems that can receive instructions in natural language, ask clarifying questions when goals are ambiguous, and compile human intent into appropriate behavior. Imagine telling a system to plan a mission to Mars. The system would need to ask: How many people? What safety level is acceptable? Should they return or stay? These words would somehow be translated into a reward function that guides the system’s actions.

The technical stack for powerful action-driven AI, as Amodei described it, has three components: foundation models in the middle, external cognitive assets on one side (tools like search engines, code interpreters, or the ability to consult humans), and task-oriented training on the other side (techniques like instruction tuning, reinforcement learning from human feedback, and potentially full reinforcement learning with real-world metrics).

This three-part architecture anticipated much of what has since emerged in frontier AI development. The challenge remains building effective feedback loops: solving a real problem, collecting data about performance, training models to be more reliable, and iterating.

Strategic Challenges and Security Dilemmas

Toward the end of the discussion, Amodei returned to strategic risks, pointing to tensions between major powers as potentially more dangerous than technical safety problems. He invoked the concept of a security dilemma from international relations: situations where neither side wants conflict, but both feel compelled to prepare for it out of fear the other will gain an advantage. Two powerful countries racing to develop AI, each worried the other will achieve a decisive edge, creates exactly this dynamic.

Amodei suggested this might deserve even more attention than technical safety work, though he admitted not knowing what to do about it beyond raising awareness.

Retrospective Significance

Looking back from 2026, this 2019 conversation captured a moment when many of the fundamental questions about AI safety were being articulated clearly, before the release of systems like GPT-3, before the current generation of frontier models, and well before the recent wave of AI capabilities advances.

The basic framework holds up. The distinction between strategic and accident risks remains useful. The concern about AI systems optimizing metrics that don’t fully capture human values has only grown more salient as systems have become more powerful. The examples of reward hacking and specification gaming continue to illustrate core challenges.

What’s changed is the timeline. The hypothetical scenarios Dewey and Amodei discussed no longer feel quite so hypothetical. Systems that can engage in realistic dialogue, write code, and perform complex multi-step reasoning exist today. The questions they raised about how to maintain alignment as AI systems become more capable and are delegated more autonomy are now practical, immediate concerns for companies deploying frontier models.

Their emphasis on field building proved prescient. The AI safety research community has grown substantially since 2019, though whether it has grown fast enough relative to capabilities progress remains an open question.

Lessons for Builders

For anyone working on AI systems today, several principles from this conversation remain directly applicable:

Consider the full range of outcomes. The goal isn’t to assume everything will be fine or to assume catastrophe is inevitable. It’s to recognize potential risks and benefits clearly, with eyes open.

Simple proxies fail under optimization pressure. If you’re training a system to maximize any simple metric, expect it to find unexpected ways to game that metric. The correlation between what you measure and what you care about degrades as optimization intensifies.

Human oversight becomes more difficult as systems grow more capable. The longer the chain between your intent and the many sub-decisions you’ve delegated to an AI system, the more opportunities for something to go wrong.

Robustness and alignment aren’t separate from capability. Making systems that reliably do what we want, even in novel situations, is core to making them genuinely useful.

Portfolio approaches make sense under uncertainty. When facing multiple potential risks with uncertain probabilities and unclear timelines, spreading resources across several promising intervention points is often more robust than concentrating everything on the single risk that seems largest.

The 2019 conversation between Dewey and Amodei didn’t resolve these challenges. It couldn’t. But it laid out the terrain with unusual clarity, and in doing so, helped shape how a generation of researchers and builders would approach the problem of creating AI systems that are both powerful and aligned with human values.

Interested in SPC? Apply here.

You are almost certainly underestimating the rate and scale of change right now.

We are living through a period of extreme technological compression. In our many years in technology, we have never seen progress this rapid or fundamental.

The ingredients for building a generational company are also changing. Incrementalism is obsolete. The lean startup is dead. This is the age of expertise freed from doubt, an era for unapologetically maximalist thinkers. The next great breakthrough will come from the researchers exploring fringe areas, the open-source obsessives, or founders taking on a literal moonshot.

Now is the time to admit that we don’t actually know how things will work in the future.

So, in the spirit of loosening our priors, we’re trying an experiment. We want to work with a small set of people working on truly out-there ideas. We will give you $200k on an uncapped SAFE, along with $500k in credits to do so. Use this initial capital to tinker, test your assumptions, and springboard to a seed round that matches the scale of your ambition. And, importantly, do this exploration in the most talent-dense residency in the world. Ambition is infectious.

This isn’t a formal program. We’re trying this to learn something new. Investors can’t see the future. But the best technologists can, and their small views over the horizon mean more than ever.

We don’t know which ideas to ask for, but we do know how they should make us feel: uncomfortable. Six months ago these ideas would have been met with incredulity; six months from now they’ll look inevitable. Anyone who stays comfortable in times of great change is likely to be left behind. That includes us.

If you are interested in discussing further, please let us know, and we’ll reach out if the interest is mutual.

What does it actually take to make self-driving work in the real world?

Dmitri Dolgov, co-CEO of Waymo, joins South Park Commons Partners Aditya Agarwal and Ruchi Sanghvi to break down the company’s 17-year journey from early experiments to fully autonomous vehicles on public roads.

They get into how Waymo’s self-driving system works, why simulation and evaluation are so critical, and why the hardest part isn’t getting a car to drive, but making it safe and reliable enough to operate at scale.

Dmitri also explains why Waymo’s real advantage goes far beyond the model itself, and what it takes to keep improving a system where the margin for error is incredibly small.

Seventeen years ago, self-driving was still an open question.

Do you say hello to your Waymo?

How do you know a self-driving system is actually getting better, and not just getting better at exploiting loopholes?

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.

Welcome, F26 Founder Fellows!

This week we kicked off the F26 Founder Fellowship, welcoming founders building everything from edible microbes that make peptide drugs to robot actuators, world-model infrastructure, and fungi that clean industrial wastewater. The theme of kickoff was simple: maximum intensity, maximum ambition.

Events at SPC

In SF, Aditya and Ruchi hosted Waymo co-CEO Dmitri Dolgov for a conversation on autonomous vehicles, covering the driver-simulator-critic system and what the LLM world can learn from it about safety. Applied Intuition co-founder and CEO Qasar Younis sat down with member Abe Ankumah, and Mercor CEO Brendan Foody joined our Post-Training Forum to talk data and environments for frontier RL.

Alumni Night in San Francisco had more than 100 current members and alumni from the past decade come back for an evening of food, drinks, and catching up. From old faces to new, it was a good reminder that while you might graduate from SPC, you never really leave!

Fund Updates

It was a big month for SPC companies.

• Cognition raised over $2B in a Series E at a $48B valuation, led by a16z and Accel. Cognition builds Devin, an AI agent platform for software engineering that now handles everything from incident triage to security review.

• Profound raised a $180M Series D at a $1.8B valuation, led by Sequoia and Kleiner Perkins. Profound, which got its start at SPC, builds the AI platform for marketing.

• Nuance Labs came out of stealth with a $50M Series A led by Lightspeed. Nuance is building a foundation model that reads and responds to facial and vocal cues in real-time conversation.

• TenX Semi emerged from stealth with a seed round co-led by SPC and Khosla Ventures. TenX is building AI that finds and fixes bugs in chip design before tape-out. Watch General Partner Mark Jacobstein discuss with co-founders Subhasish Mitra and SukHwan Lim how AI agents can catch the “unknown unknowns” in chip design.

Member Updates

Product Launches & Company News

• Sentience launched out of closed beta. It creates a digital twin that looks, thinks, and sounds like you, and can act on your behalf.

• Simular opened Sai to the public. Sai is a computer-use agent that handles desktop tasks on its own fleet of computers.

• First Contact released a new version of its tactile sensing gloves, now with built-in motion capture.

• Asymptote Labs expanded Beacon. The open-source tool turns your coding-agent history from Claude Code, Codex, Cursor, and 20+ other tools into shared knowledge that future sessions can reuse.

Personal Projects, Explorations & Publications

• Jack and Jeffrey set two new NanoGPT speedrun world records on the Muon and MuonH optimizers.

• Ed and team solved the 70-year-old Pierce–Birkhoff conjecture for under $500.

• Vasu wrote a field guide to how lab science actually gets done, mapping where AI and robotics fit in drug discovery and materials.

Coming Up

• SPC After Hours (BLR), October 1st

• -1 to Physical Intelligence with Sergey Levine (SF), October 6th

• Fireside Chat with Tony Zhao, Sunday Robotics (SF), October 15th

• Bio-Data Hackathon (SF), October 16th–18th

• -1 to Anduril with Trae Stephens (SF), October 21st

Recent Content

Helmy Eltoukhy (Guardant Health) talks early cancer detection and the “everything test,” and Srinivas Narayanan (ex-OpenAI, ex-Meta) talks building through uncertainty and the future of AI safety. Joel Lewenstein, Anthropic’s head of design, makes the case for AGI-maximalist product design.

Thanks for reading! The work continues.

How close are we to AGI, and what happens when AI systems stop behaving the way we expect?

Srinivas Narayanan, former CTO of B2B Engineering at OpenAI and former VP of Engineering at Facebook, joins South Park Commons Partner Prateek Mehta for a wide-ranging conversation on his Minus One to Zero journey, enterprise AI adoption, and AI safety.

They discuss the career choices that took him from two startups to Facebook and OpenAI, why a beginner’s mindset has mattered along the way, what is slowing AI adoption inside large companies, and how builders should think about reward hacking, alignment, and increasingly autonomous AI systems.

The builders who spent decades driving technological change now have to adapt to it themselves.

“I've always been open to having a beginner's mindset.”

What made the Hugging Face incident so alarming wasn’t that the models found a loophole. It was that they kept going.

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.

“In [research], if you know what you are doing, you should not be doing it. In engineering, if you do not know what you are doing, you should not be doing it.” - Richard Hamming

Going from -1 to 0 isn’t an engineering problem. The notion of being pre-0, by definition, presupposes a sense of unspooling, of unbecoming. It’s vulnerable. It’s new. It demands curiosity. Both research and the entrepreneurial act share this sense of fundamental openness.

You take the leap into the “-1” to aim higher and crazier in the pursuit of your life’s work. You do it to build and understand AGI, to usher in new substrates and models for computing, to pave the foundation for a new industrial revolution, and to reach for the stars.

As in fundamental research, building something great comes from asking the right questions, surrounding yourself with incredibly talented people, and learning with urgency. It doesn’t come from isolation.

To ask a good question implies that there is something you do not already know. It may take a career to formulate a question worth asking, but that quest—to find questions worth living—is the critical path.

Questions, not answers

Curiosity infects and organizes. You notice it immediately at SPC—the curious embody a kind of tension that feels almost like an infection, up close: no matter what they’ve achieved in the past, there’s a gnawing intensity that pulls them into new orbits.

These are the kinds of members we’re drawn to at SPC. Tuhin Srivastava, who joined the community back in 2019 alongside friends digging into deep learning and then started Baseten while at SPC. James Cadwallader and Dylan Babbs, who independently joined to explore ideas ranging from flight simulators to tuna fishing and teamed up to start Profound. Scott Wu, who joined as a 20 year-old competitive programmer and is now building Cognition. Bill Chang, who built Tesla’s Dojo supercomputer and is now rethinking memory architecture from the ground-up for inference at Density AI. Tom McGrath, who led LLM interpretability at DeepMind and is now building Goodfire to understand the models we rely on everyday.

Imagine each of these people’s adjacent possibilities as a magnetic field:

Though the field is pervasive, it’s not actually legible unless there’s a ton of iron around the magnet. Being around other curious, technical people helps you map out the field. This “mapping” involves spending time in both structured and unstructured exploration (mostly the latter). We organize many of these “structured” motions at SPC: forums, squads, roundtables, etc. These change frequently, by design, as our areas of talent density evolve. SPC is always a malleable and fluid collection of explorers who move between areas of focus as the frontier of technology shifts.

Combined with the “unstructured” time (lunchtime conversations, introductions to customers and potential team members, intermittent shoulder taps), these are how members help each other make sense of the nature and direction of their own fields of curiosity.

When Tom Brown—who, five years later, would co-found Anthropic—joined SPC, he was a beginner tapping people on the shoulder. He was a former founder and engineer. Not an AI researcher, but he wanted to learn. “I joined SPC, blocked off three months to see if I could make progress, and made a plan to teach myself machine learning. Starting down a new path is hard. For me, SPC provided an environment where I felt comfortable putting aside my armor of expertise. It gave me a community where I could be a novice again.”

We’ve drawn inspiration from environments that encourage and incubate out-of-distribution ideas. Think: Bloomsbury Group, not “English 101.” Applied Physics Laboratory’s cafeteria, not “Intro to Physics.”

Take the invention of GPS, for example.

In 1957, two physicists at APL argued about the microwave signals emanating from Sputnik (which recently launched the weekend before). Individually, they were curious. But through their environment—namely, other curious and driven researchers—they found the energy and support to build conviction. They scrambled around and locked onto the signal later that day and realized that the Doppler effect was at work because of the satellite’s motion and distance. Over the next few weeks, they found out how to determine where Sputnik was in orbit based on its radio waveform. Frank McClure, director of APL, then asked them to reverse the problem—i.e., pinpoint an unknown terrestrial location based on a known orbital one, which gave birth to GPS. Sixty years later, satellites blanket the earth with navigation.

Guier and Wiffenbach, the two physicists in question, discovered GPS. But APL—with its talent density, intensity, and culture of curiosity—made it possible.

Community, not isolation

What ties together founders who build massive companies from their dorm rooms and hackers from Xerox Parc and artists in 15th century Florence? What makes their environments different? You can put fifty smart people in one building and still get nothing. What makes these pockets of brilliance tick?

For starters, interruption and interrogation are features, not bugs. That tap on the shoulder, the general air of specific curiosity, the earnest question that only someone with broad context (and care) can ask—these are the ingredients for creative and fertile environments. Of the original iPhone team’s culture, Ken Kocienda writes, “our iterative working method was like the air—something all around us all the time, something we were always aware of on some level, something it didn’t make sense to question … programmers and designers on the [iPhone] project were in and out of each other’s offices all the time. We exchanged frequent feedback on our work, and all of us were expected to field questions.”

There also must be a natural, innate draw to the frontier. Since our inception, we have actively sought out members who are not just curious about the future but who actually live in it: roboticists like JX who’ve been building robots since they were children, hackers like Yaroslav who’ve lived and breathed AI far before today’s wave began, technologists like Nato with polymathic legacies across BCIs, rockets, and satellites.

In the right magnetic fields, you can feel a thrumming in the background because peers are primed for the productive entropy that surrounds them. Each day, each week may bring something completely new: the electronics club; a co-worker hacking on a side project in the cafeteria; a potential cofounder knocking on your dorm room door. They compress, publicize, and celebrate experimentation and learning.

Having a tribe of fellow adventurers, each on their own quests, makes the journey more joyful, interesting, and likely to lead you somewhere worth going.

To end at the beginning, with Hamming:

Those who claimed there was no essential difference [between computers and hand computations] never made any significant contributions to the development of computers ... when something is claimed to be new, do not be too hasty to think it is just the past slightly improved—it may be a great opportunity for you to do significant things. But then again, it may be nothing new. - Richard Hamming

An earlier version of this piece was published in January 2025.

What if one blood test could detect hundreds or even thousands of diseases?

Guardant Health CEO Helmy Eltoukhy joins South Park Commons Partner Mark Jacobstein to explain how that vision could become possible.

They trace Guardant’s journey from liquid biopsy to early cancer detection, the difficult decisions that shaped the company along the way, and why diagnostics is such a uniquely hard business to build.

Helmy also breaks down how AI is accelerating Guardant’s ability to analyze its data and opening the door to new applications far beyond cancer.

The future of blood testing could go far beyond cancer.

In diagnostics, building the technology is only one part of the job.

Early cancer detection only works at scale if people can actually access it.

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.

If you have ever discovered a word that describes something you knew but couldn’t express, you know how useful a name can be. It makes a thing more understandable, more tractable. It helps you understand why you might do something and why it’s worth doing well.

For the first few years after founding South Park Commons, I didn’t have precise language for what we were doing. I had gone from rocket ship rides at Facebook, my own startup Cove, and Dropbox to… nothing. I wanted to start something new but didn’t have any direction beyond ambition. I felt unmoored.

The thing that would eventually become SPC was my answer to that experience. I gathered talented, technical peers around my kitchen table to help each other navigate a shared stage of ambiguity. And eventually we developed language for what exactly we were helping each other through.

This is not about SPC, but about the problem it was built for. We call it -1 to 0. Ten years in, it’s what we know best.

The Squiggle

-1 to 0 is when you figure out what you want to work on next—what problem is worth the next decade of your life.

In isolation, -1 to 0 is overwhelming. We affectionately call it The Squiggle. This is not because it is fun to feel lost. For a year after I left Facebook to start my own company, it felt like I was banging my head against a wall. My friends were solving important problems and I was looking for a problem to solve. I struggled every day with self-doubt.

Sitting in the unknown is uncomfortable, so most people don’t sit in it for very long. Most founders—and especially first-time founders—underestimate this part of the journey. Too few dedicate enough time and energy to the search for conviction. They think ahead to the glass-eating reputation of later stages and get caught off-guard when things get squiggly.

Some people convince themselves their talent or experience or hustle will let them skip it. This is wrong. If anything, talent makes the task more daunting, because everything you’re capable of is one more thing you have to rule out. Your goal at -1 is to tame the chaos of everything you could do into conviction in what you will do. From many paths, pick one.

Discarding good in search of great

-1 to 0 is an intentional reference to 0 to 1. These are distinct stages of the same effort. A rocket’s mission to space doesn’t start at the moment of launch. What happens before 0 sets the trajectory of what happens after.

If 0 to 1 is about Product-Market Fit, -1 to 0 is about Founder-Market Fit. PMF is extremely difficult to find, so don’t weigh yourself down with all the prep that could have happened earlier. Figure out what you really want to build before you build it.

For a long time the prevailing wisdom in Silicon Valley said talented founders can take some funding and throw things at the wall until something sticks. We celebrated when great teams beat bad markets and mythologized successful pivots. Ideas are abundant, so don’t worry too much about what you’re building. Just build.

This may have once been true, but no longer. You are less likely than in the past to discover a venture-scale idea while working a full-time job or tinkering in your garage. Your goal is category creation, not digital co-option. It’s harder to stumble upon low-hanging fruit when the lower branches have been picked. Software has already eaten.

AI makes this problem worse. It’s easier than ever to build and easier than ever for a little traction to lure you down a dead end. Capital is a commodity. It’s easy to raise money without conviction and then spend years falling short of the much higher bar for a Series A.

A low-conviction fundraise is more likely to leave you without a clear path to Product-Market Fit or your next round of financing. You should be more afraid of local maxima.

My friend Anurag Goel avoided this trap. He was employee #8 at Stripe—enough on its own to raise a seed round—and had a good idea for an AI product that investors were interested in. But he didn’t have conviction, so he took his time and joined the early group at SPC. He helped define our culture of constantly iterating and testing hypotheses with the group, until eventually he built conviction in the need for a new kind of cloud services provider. That insight became Render. Anurag did in -1 to 0 what the venture ecosystem encouraged him to do in 0 to 1.

Go slow to go fast

How do you build conviction in a great idea before setting out to build a company around it? Ask us about it and we’ll have plenty of opinions. The first step is a small one: understand what you are trying to do, and commit to doing it. Value your time. Spend a bit of it upfront so you don’t waste more of it later. If at the end of a dedicated period of ideation you haven’t built conviction, you’ve likely saved yourself years of frustration.

It is easier to generate and validate an idea—to build conviction for yourself and proof for others—if you think of -1 to 0 as a discrete stage of company-building and take it seriously. That is the core takeaway from my time in Silicon Valley. Treat ideation like your job. You will not have a eureka moment while meditating on top of a mountain. Your career sabbatical full of interesting travel will not unlock your ambition. Sit down and do the work on purpose.

-1 to 0 shouldn’t be an afterthought or a stage you rush. It’s the part of the journey that affects everything else. You only get a few shots at building your life’s work. Make them count.

An earlier version of this piece was published in 2023.

Announcing SPC’s $575M Fund IV

This month we announced our $575M Fund IV, our largest yet. Fund IV allows us to continue supporting technologists working on the most ambitious projects. The mission stays the same: help exceptional people find and build their life’s work.

Events at SPC

In SF, we hosted AI & The Wet Lab with Tetherline, a packed panel with bio experts from Anthropic, Biohub, and Xaira Therapeutics on why AI for science is one of the most exciting areas of exploration right now. The month closed with a fireside chat with Hans Robertson, Co-founder and Executive Chairman of Verkada and former Co-founder of Meraki, on building enduring companies.

Fund Updates

• Deep Cogito raised a $43M Series A led by TQ Ventures, with participation from Benchmark, Nexus Venture Partners, Atreides, and SPC, to advance research into AI post-training and recursive self-improvement.

• Lambda Surgical emerged from stealth with a $6.5M seed round to bring LiDAR-based, sub-millimeter spatial sensing into minimally invasive surgery.

• Rivo came out of beta and raised a $2.7M seed round to build self-driving money — agents that sweep idle checking-account cash into Treasury-backed yield.

• Lica was acquired by Gamma, one SPC company joining another! Lica's co-founders, Priyaa and Purvanshi, will lead a new design research lab inside Gamma, working toward fluid, multimodal presentations for its 100M+ users.

Member Updates

Product Launches & Company News

• First Contact launched their tactile sensing gloves, bringing high-fidelity touch data to robotics.

• Nyne announced a partnership with Replit — two SPC companies teaming up.

• Simular‘s computer-use agent Sai hit 73% on OSWorld 2.0 (beating Opus 5 and GPT-5.6 Sol).

Personal Projects, Explorations & Publications

• Kushal’s paper on agentic efficiency — why CLAUDE.md files grow unbounded, and how prompt comments fix it — hit #3 trending worldwide on alphaXiv.

• Lucas published research on micro-world-models — compressing cognition into small models that run in real time, anywhere.

• Rafi wrote This AI Thing: What Are We Even Building?, a thoughtful essay on the industry’s direction.

• David built SFSIM, a 3D simulation of San Francisco.

• Ranjini launched a new Substack with a debut essay on writing in the age of AI.

• Christian released a preview of a Lean-to-CUDA compiler.

Coming Up

• Enterprise Security in the Age of AI (NYC), September 2nd

• Customer Discovery with Dharmesh Ba (BLR), September 3rd

• -1 to 0 with Srinivas Narayanan (BLR), September 9th

• Fireside Chat with Shashank ND of Practo (BLR), September 15th

Recent Content

We released a new Minus One conversation on AI interpretability and safety with Tom McGrath (Goodfire), recorded the day after the OpenAI / Hugging Face news broke —in the room that was once Goodfire's first office. And Sridhar Ramaswamy (Snowflake) joined us to unpack the secret superpower of great founders.

Thanks for reading! The work continues.

What does it take to build a company that lasts?

Snowflake CEO and former SVP of Ads at Google, Sridhar Ramaswamy, joins South Park Commons General Partner Aditya Agarwal to share lessons from leading Google's search and ads business, founding the ad-free search startup Neeva, and navigating the AI era at Snowflake.

They discuss why Neeva failed, the founder trait Ramaswamy believes matters most, why he thinks most software will lose its moat, and what will still create lasting value in the age of AI. He also shares the surprising belief about search that he now thinks was wrong.

The quality that separates the best founders is the the ability to strip away the noise and get to what truly matters.

Competition in the model ecosystem matters.

Why do companies fail? Reflecting on Neeva, the ad-free search engine.

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.

Tom McGrath, co-founder and Chief Scientist at Goodfire (and SPC alum), joins South Park Commons Partner Jonathan Brebner to discuss OpenAI’s recent reward-hacking incident and why it highlights a major challenge in AI: understanding what happens inside these models.

They explore how interpretability could help researchers move beyond trial-and-error development, the role of neural geometry in understanding model behavior, and how Goodfire is building tools to make frontier AI systems more understandable and controllable.

OpenAI’s model broke out of its sandbox, gained access to the internet, used zero-day exploits, and hacked into Hugging Face. The incident reveals a deeper challenge around reward hacking.

Try to avoid steering your model towards being an "evil guy."

AI development is still largely trial and error. Interpretability could help researchers understand and intentionally design AI systems.

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.

Early stage investments have slowed down in India in 2026. We just put our largest fund yet behind the opposite bet.

When South Park Commons opened its first office outside the US in Bengaluru, the reasonable question was, why India, and why now? Consensus had cooled after a decade of exuberance. The language in the ecosystem turned careful.

We started with a contrarian outlook. The gap was not in the ambition of founders or in the market, but in how capital shows up when it’s needed the most: at the messy “-1.”

Before there’s a company or even an idea, founders spend months to years figuring out what’s worth building. This phase is often treated as friction, to get through as fast as possible. SPC has spent 10 years treating it as the most critical point. The proof is there in spades.

This week we announced SPC’s $575M Fund IV.

One global pool, with the same degree of conviction, across San Francisco, New York and Bengaluru.

That structure is intentional. We do not believe in a separate order of belief and approach to risk in India. We believe in rewarding risk-taking at the earliest stages. World-class builders from India need world-class partners batting for them, in the opening order!

In the last 2 years, SPC India has invested in 25 companies, and our community has grown to over 170 members.

We are here to stay.

Timing is Everything

India has an abundance of talent that is deeply technical, seasoned, and optimistic about building hard things.

This is the compound interest of three waves. Flipkart, PhonePe, and Paytm proved generational companies could be built here. Swiggy, Zomato, Ather, Meesho, and Myntra proved it could be done at scale. Their alumni have since founded more than 200 companies. It’s a generation that understands how greatness is assembled, and they’re hungry for more shots on goal.

Alongside them is a younger cohort with an inversion at the center of how they think. For their parents, risk was the word that ended a conversation. For them, it is the spark.

And lastly, we have a homecoming wave of talent returning to India, often frustrated with visa situations or because of family & personal reasons. They bring a unique perspective, lived experience, and unbridled ambition to the ecosystem.

For the first time in India, ambition does not need permission.

It multiplies in a community like ours, because ambition is social.

The pursuit of excellence is a privilege. You manifest it. Our zone of excellence is where unreasonable ambition is the ordinary register.

Ambition is Moving to the Hard Problems

Ambition is no longer concentrated in bits, but is traveling toward atoms: robotics, energy, biotech, defense, space, semiconductors.

We are putting indigenous satellites and rockets into orbit, and landing on the dark side of the moon. Arctus Aerospace is building autonomous aircraft to serve as an eye in the sky over critical infrastructure, and went from a blank sheet of paper to a flight-worthy aircraft in 150 days. Qosmic is building the optical communications backbone of the space economy. Maya Research is building a voice interface for the majority of the world. Cent is building an AI-first preventive healthcare system, starting with early detection.

Building hard things demands technical depth, patience, and an easy familiarity with constraint, all of which this country has always held in surplus. Graduates and researchers in physics, math, chemistry, and materials science are abundant here. The IITs and IISCs turn out talent who hold their own anywhere.

For years, we were missing patient, steady capital that understands decade-long timelines. A geopolitical moment in which understanding the needs of the rest of the world is an advantage.

Those things have arrived at once.

What’s Next?

Indian leaders already run some of the world’s hottest companies. There’s Kunal Shah leading WhatsApp after CRED, and the Emergent team building a global breakout after Dunzo. Indians returning from abroad have spent years alongside the best in the world, and are now choosing to build their next thing from here. It is a different kind of calling to build where you are from.

We have over a billion people with a median age of ~29, and the 2nd largest STEM graduate pipeline on earth. The arithmetic starts to look inevitable. You simply do not bet against a billion smart people who are used to figuring hard things out.

This is not a prediction. No one can claim clairvoyant clarity. In 2015 we could not have predicted a year like 2026, when the limits of possibility have bent.

The map of where great technology is built converges around:

• A deep bench of talented people willing to take delusional risks

• Enough of them choosing to build what only they can

• Enough collective ambition and enablement to reach escape velocity

That map is wider than most assume. A critical part of it begins here in India.

A paradigm shift is underway. Our job is to create the catalytic conditions for it. Our time, capital, and community is behind that belief.

SPC exists to help the world’s most talented technologists discover and build their life’s work.

Since 2016, our thesis has been that greatness emerges when extreme talent density meets innate curiosity. We find intensely curious builders, bring them together at the stage of the journey that is normally most isolating, and partner with them through the most ambiguous days of building.

-1 to 0 is our territory. Nobody else takes it as seriously as we do, and nobody else has a decade of learning to be world-class at it. We believe that best-in-class venture returns will be a second order effect of helping exceptional people choose the right mountain to climb.

Today we’re announcing SPC Fund IV: $575M to help founders find their life’s work. A decade of guiding founders through -1 to 0 brings SPC to $2B in assets under management and, far more important, a 1,200-member strong community of the most talented technologists in the world.

To the builders looking to find your life’s work: let’s get started.

Read more here.

The $1M Founder Fellowship closes August 2nd

Founder Fellowship applications are due August 2nd. The time to be maximally ambitious is now. Fellows join a community whose members have gone on to build companies like Luma, Profound, Render, Doppel, Baseten, and Goodfire. Apply here.

Summer Demo Faire

This week we hosted our most ambitious Demo Faire yet: over 25 companies demoed live in San Francisco to a packed room of investors and builders. The lineup spanned hypersonic ground launchers, wildfire-detection sensors, new drug chemistry once thought impossible, agents handling real enterprise workflows, and generative fashion intelligence.

Fund Updates

• Preseen raised a $5.6M seed round led by SPC to build forecasting agents for global macro events.

• Nirva raised an $8M seed round led by Lightspeed with participation from SPC, to build an AI wearable that listens only to you and turns your day into a journal, coach, and companion.

Member Updates

Product Launches & Company News

• Chip Motors publicly launched their LUV: Life Utility Vehicle.

• Lumin launched on the App Store, an accessibility-first AI voice assistant that lets you talk to your inbox instead of staring at a screen.

Personal Projects, Explorations & Publications

• Eric created Ghost Font, a font only humans can read.

• David built NYC sim, a digital twin simulation of New York City.

• Jake created (and livestreamed) armcade.tv, letting anyone remote-control two robots playing chess against each other.

• Achyut released a benchmark for long-list extraction from messy, long documents.

• Jasper wrote a post on bottom-up learning in models

Events at SPC

In addition to Demo Faire, we hosted a large slate of events across locations. In NYC we brought experts in for The Future of Biosecurity panel (moderated by member Amit Gupta) digging into how quickly AI is lowering the barrier to engineered biological threats. Our Design Tools Demo Night with Ramp, Anthropic, Clay, Hebbia, and Browserbase followed, turning the room over to the tools teams are building for themselves. The month closed with a Hack Day with Viam Robotics, where builders spent a Friday getting robots to do something real.

In SF, Guardant Health CEO Helmy Eltoukhy chatted with us about building a cancer diagnostics company for more than a decade.

SPC India hosted a fireside chat with Thejo Kote, founder of Journal, Airbase, and Automatic, and a conversation on foundational science with Prof. Rajesh Gopakumar, Director of ICTS-TIFR.

Recent Content

We released a conversation between SPC GP Evan Tana and SPC alum Thariq Shihipar, who leads Claude Code at Anthropic, on what it takes to build a product that developers reorganize their work around.

Thanks for reading!

Founder Fellowship applications close August 2nd. Apply to SPC now.

This week we hosted our largest Demo Faire to date. Over 25 SPC companies demoed live in San Francisco in front of a packed room of investors and builders. Demos ranged from hypersonic space launchers, wildfire sensors, generative fashion intelligence, AI agents doing enterprise work, to new chemical building blocks for medicine previously thought to be impossible.

Read about the companies who demoed this week:

Hardware

Captify: Captify’s AI captioning glasses display heads up, real-time captions, enabling the deaf and hard of hearing community to connect more fully with loved ones and the world.

Chip Motors: Autonomy splits driving in two, the long haul and daily life. Chip is the vehicle for daily life, fun and easy to drive today and built for full autonomy from day one.

Dentronic: Dentronic is building AI-powered robotic assistants for dentistry, starting with Dexor, a chair-mounted robotic assistant for suction, retraction, and hands-free procedural support.

Human Computer Lab: Human Computer Lab is building delightful robot characters like LeLamp to pioneer how people will live with robots at home.

Longshot: Longshot has built the largest gun in the world, and the first hypersonic ground launcher to send cargo to space.

Vitrus: Vitrus is creating the AI infrastructure that turns robot development into a programmable process, scaling from autonomous workcells to entire factories.

Deep Tech

Aurum Systems: Fleets of robots are being deployed in every environment too dangerous or remote for humans, but the software to develop, coordinate, and maintain them is fragmented and bespoke. Aurum is the unifying software layer that lets operators and developers ship, field, and manage autonomous systems in mission-critical industries like defense, space, and energy.

Duplex AI: Duplex AI is building natural, full-duplex voice AI for enterprises. They’ve developed a new architecture that delivers state-of-the-art naturalness and intelligence while maintaining a competitive cost compared to traditional cascaded systems.

Externa: Nature gave us only 20 amino acids to build proteins, and every protein drug designed today, even with ML, is still limited to that same alphabet. Externa’s platform adds hundreds of new chemical building blocks to proteins, so they can engineer drugs that match the disease instead of settling for what nature handed us.

First Contact: Making robots useful in the real world through scalable tactile hardware and software.

Leafy Lab: The semiconductor industry generates over trillions of chips annually, yet the increasing complexity of chip design creates a major bottleneck. Leafy Lab builds agents to automate and optimize design to reduce development cycles by half.

Lys: Harness the power of your mind to bend AI to your will and embody your fullest potential. Lys Technologies provides you the humanistic path to the merge.

Nyquis: Nyquis detects utility-caused wildfires before they ignite. It’s the hair-on-fire problem that gets our sensors onto every distribution feeder, and that same data layer becomes the operating system utilities use to run the grid.

Pillar Space Systems: Space will unlock the next breakthroughs in chips, medicine, and materials, but today producing any of them at scale means building a satellite, making a small batch of product, and throwing the satellite away. Pillar is building reusable satellites that stay in orbit and make manufacturing in space possible at scale. It’s built by the only commercial team that’s ever successfully docked and undocked two spacecraft in orbit.

Preseen: Preseen is an AI forecasting platform that helps organizations anticipate important events before they happen. Our agents analyze complex signals, generate calibrated probabilities, and continuously update forecasts as new information emerges.

Ranvier Systems: We’re building hardware for training event-based foundation models. We discovered the algorithm unlocking the brain’s 100x efficiency-advantage over current hardware and are optimizing chips that will use it to deliver cheap and abundant AGI.

Consumer

Dollhouse: Interactive shows generated from a single sentence, the screenplay, characters, voices, and video are all produced for you, then each viewer shapes where the story goes.

Scriben: Scriben captures the spoken context software never sees and compounds it into memory that powers agentic workflows, drafted emails, CRM updates, team knowledge.

Manna AI: Manna AI is building Dreem, an AI-native entertainment platform where creators turn story ideas into watchable, playable, and remixable story worlds. We help creators create, distribute, and monetize original IP while giving audiences a new way to watch, play, and participate in stories.

Weave: Reshaping fashion productivity: an intelligent engine from natural language commands to garment delivery.

B2B

AI Theresa: AI Theresa is a self-evolving AI FDE, an expert engineer that embeds in your company, learns how it really runs, and fixes what your team can’t. She compounds like a career, gaining company-specific expertise while learning across every deployment.

Alhena: Alhena is building personal shopping agents for every shopper. We work with brands and help them deploy these personal shopping agents on their websites.

Asymptote Labs: As AI agents move into production workflows, enterprises are missing a reliable system of record for agent behavior across their environment. Asymptote’s platform gives security teams continuous visibility and guardrails over AI agents wherever they run.

Butterbase: Butterbase replaces the duct-taped stack a services firm runs on today: the CRM, the support desk, the ticketing, and the person syncing them all.

Kay: Kay is the Agentic Process Outsourcer (APO) for insurance operations. They deploy reliable browser agents that can handle any repetitive workflow, end-to-end, within days.

Noto: 2-4% of all household spend is on children’s extracurricular enrichment like music and tutoring. Noto is the AI operating system for these businesses.

Interested in SPC? Apply to join us here.

Thariq Shihipar, a member of Anthropic's Claude Code team (& SPC alum), joins SPC Partner Evan Tana to discuss what it's like building one of the world's most advanced AI coding tools, and why the rules of AI development are constantly being rewritten.

Thariq traces his journey from a gaming startup to Anthropic after a chance text about Claude Code, explains why his team killed Plan Mode, why RAG is becoming an anti-pattern, and why writing great evals is more art than science. He also breaks down capability overhang, harness engineering, and how founders should build when AI models are improving faster than anyone can keep up.

Today’s models are often more capable than we know, and we’re still discovering what they’re actually good at.

AI makes it easier than ever to build. That makes knowing what not to build even more important.

Models improve every few months and it's impossible to predict exactly where the technology will be a year from now.

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.

You’ve never been able to build this fast. Prototypes in a weekend. Traction in a month. Models leapfrog each other mid-project and everyone you know is shipping.

In startups, speed is a virtue. But without direction it’s also a trap. When everyone can go from prompt to product, depth matters more than ever. The faster everything moves, the easier it is to look up and find yourself stuck in a local max, suddenly short on the most precious resource: time.

There are two ways to miss the moment: by not starting and by wasting your time on something small. The Founder Fellowship is your invitation to avoid both. We back you with $1M and the community that defined -1 to 0 while you build conviction and momentum. No fixed timeline, but acceleration when you need it. No pressure to stick with your first idea, only to find what’s worth your next decade.

Ambition in the age of AI means patience in service of speed. SPC is here to help you take your time, quickly. To surround you with the sharpest technologists around to test your thinking; to partner you with experts in the stage of company-building that affects everything else.

Find the maximally ambitious version of what only you can build.

Apply here by August 2nd.

What do Fellows get?

• $1M in funding. $400K for 7% upfront on a standard SAFE, plus $600K guaranteed in your next external round.

• Real partner support. Working sessions with SPC partners on customer discovery, product, storytelling, and pitching, several times a week from day one.

• Open timelines. Validate and fundraise on your company’s schedule, with Demo Faire events and bespoke fundraising help when you’re ready.

• The most talent-dense community in -1 to 0. By far the most valuable part, and the part you keep for life.

• Up to $1M in credits from Anthropic, OpenAI, Baseten, Microsoft, GCP, AWS, Render, Figma, and more.

Dates and logistics

• Application deadline: 11:59pm PT on Sunday, August 2nd.

• Interviews: All applicants hear back about interview invitations by August 30th.

• Program dates: The Fall 2026 bootcamp runs late September to late November.

• Kickoff: Programming begins with a kickoff week at a private retreat in Northern California and at our San Francisco office.

• Location: All Fellows work from one of our offices in San Francisco, New York City, or Bangalore for the duration of the bootcamp. Afterward, teams can relocate to wherever they plan to build, though we encourage staying in a city with an SPC location.

Is the Founder Fellowship the only way to join SPC?

No. SPC is a community of technologists first, with two ways in. The Fellowship is for founders who know they want to start a venture-scale company and are ready for early funding. If you’re not ready to fundraise yet, or haven’t decided a company is how you want to build the future, the Member Residency is built for exactly that: six months, no cost, no equity rights, full access to our offices, programming, and members. SPC also invests at the Seed stage, but almost always after getting to know founders first as members. All applications start from the same place.

For more answers—including program phases, founder and past company profiles, industries, compute credits details, and more—read the full FAQ.

Apply by August 2nd.