Since then, it's become a defining characteristic for how USV assesses startup opportunities. In many ways, every thesis USV has ever articulated has been built upon this idea. You can trace it through direct-to-learner education (Duolingo), programmable money (Coinbase), and publishing without publishers (Twitter). Our whole quest has been how to overhaul market structures instead of simply creating efficiencies within them. But only recently did it become clear to us how AI makes the potential for true obliteration of existing business models and markets that much stronger. In other words: there’s never been a better time to obliterate. 

So what changed? 

The internet collapsed the cost of distribution and stripped the market power from the media companies that controlled it. AI does the same thing to expertise. As expert-level intelligence becomes abundant to anyone with an agent, the people, companies, and institutions whose power came from gatekeeping that expertise start to lose it. Instead of renting expertise from a gatekeeper, you can access and deploy it directly. No gatekeeper required. That shift, from the supply side to the demand side, is what actually restructures a market. We've already made a few investments that we think have the potential to obliterate the markets in which they exist, thanks to AI.

Doctronic, as one example, doesn't simply help streamline the experience between doctors and patients. Instead, it democratizes medicine by putting an AI doctor (that can even legally write prescriptions, starting in Utah) into the pocket of anyone with a smartphone.

Isembard doesn't just sell software to existing factories. It rebuilds the factory itself by packaging the expertise of running a high-precision manufacturing line into software. The industrial capability that used to live in a handful of aging machine shops becomes something you can deploy through AI.

Or Cofounder, a company that doesn't make product development, sales, marketing, and distribution easier, but launches and manages full businesses on behalf of its customers.

In each of these examples, if they work, the markets they're in won't look the same for much longer. And in nearly every category of business, across both consumer and enterprise, it's clearer than ever that AI has the potential to reset the table completely. Almost everywhere you look, there's an opportunity to transfer the balance of power away from a gatekeeper and towards a customer. The opening is that big.

This is what we mean by obliterate.

To advance from here, we face a dual challenge: we need to construct new infrastructure, such as massive solar arrays, and we need to maintain the existing infrastructure, much of which is crumbling. As Stewart Brand beautifully captured it in his most recent book The Maintenance of Everything, maintenance is a civilizational challenge.

Thankfully, we are in the midst of a robotics revolution. Robots are going from highly repetitive tasks in precisely controlled factory environments to being useful in complex outdoor conditions where most of our infrastructure lives.

We are excited to announce our investment in Gritt, which is pursuing the infrastructure robotics challenge, starting with the build-out of grid-scale solar arrays. Gritt has already deployed their robots in the construction of tens of megawatts (a number that grows by the day) and is currently contracted for another 2.8 gigawatts with leading US contractors. Co-founders Puneet Puri and Vishal Dugar’s vision for Gritt is to build the world’s infrastructure. You can read more about their plans for Gritt on their blog. Gritt is hiring!

Where prediction markets are today is no small feat. After decades of interest in the concept and several failed startup attempts, many had written off the category as too challenging or still too early. Yet, two teams were patient and relentless enough to bring these markets to life. Kalshi fought through years of regulatory morass to offer event contracts to US users and has meaningfully reshaped sports betting. Polymarket was equally pioneering and has carved out a large user base globally.

Prediction Markets' Second Wave

While we are large fans of the work these companies have done to create the category, we also strongly believe that prediction markets are still in their early innings. With a modern toolset, we believe a much richer product with a stronger foundation can be built today than what was available to the leading platforms six years ago. As the category expands, there will be room for multiple large players.

The last 15 years of observing and investing in novel market infrastructure and exchanges has also informed our view. Consistently, much later entrants with a novel tack on the market have been able to take serious market share. Like what Binance did to Coinbase, or Blur did to OpenSea, or FTX initially did to Binance, or Morpho did to Aave, or Hyperliquid did to dYdX – and so on. One common theme in these examples has been focusing on a key underserved user segment.

An Exchange Built for Traders

When we met Ivo and Matt, they had a very simple but important observation: real traders are still not well-served in prediction markets. They’d both spent meaningful time as full-time traders working across different markets, venues, and scales of operations, and then worked together at dYdX where they helped shape one of the most innovative exchanges of its era. They have deep experience across all facets of an exchange, and when they started exploring prediction markets, they quickly felt like the existing venues had none of the feel and ergonomics that very mature exchanges have.

Everything from order types, to execution reliability, to capital efficiency, to fee structures still felt quite early. What emerged from this experimentation is Pascal, a new prediction market venue built entirely by traders for traders. It is an exchange built from the ground up to meet the needs of serious participants – not just an expansive feature set, but also a great interface and a highly performant and reliable API, all without custodying users’ capital. In many ways, it is the type of venue that Pascal's team of former-traders would be excited to use.

The Series A

We are very excited to have led Pascal’s Series A financing. Pascal came out of stealth into private beta three weeks ago. They’ve received very positive feedback on UI, API quality and design, and liquidity and have already processed over 2 million contracts – all organic volume with fees and without incentives.

If you are an active trader on prediction markets, I highly recommend giving the interface or API a try. They already have great liquidity across key markets and are just getting started. They have built an exceptional, lean team in New York and are hiring, including for frontend engineers.

Because the space is so vast, humanity to date has synthesized a tiny fraction of all possible materials. Much of what works so far is the result of lucky accidents combined with brute force exploration. Thankfully we now have AI to assist us. One way to think about the superpower of AI is that it can discover structures in high dimensional spaces. For us humans the world is only directly accessible through the 4 dimensions of our perception, which makes imagining and discovering structures in higher dimensions really challenging. AIs do not face this limitation. 

The team at alqem is setting out to use this power of AI to find magnetic materials that do not need rare earths to be powerful. With most transportation going electric and billions of robots on the way, the demand for electric motors and actuators is exploding. Figuring out how to make the necessary magnets from cheap and abundant elements will be a major unlock.

The alqem team has a great background for taking on this challenge. They built the largest open materials data set before embarking on this mission and have pulled in leading researchers in the field. The AI models they are building propose candidate materials, the team synthesizes these, and the results, most of which no one else has ever measured, feed back to make the next round of predictions sharper. This loop generates proprietary data which builds a compounding advantage as the models improve.

You can find a lot more detail on them and their approach in their own post. We are excited to be backing them in this quest. Alqem is actively hiring.

You can read their announcement here.

What began as a simple "chat app" experience has matured into the agentic era. More powerful models, combined with better tooling to orchestrate and connect them, have made it possible to build truly capable agents in many forms. Coding agents are just the tip of the iceberg. This shift favors a different kind of architecture. A chatbot can be vertically integrated: one model, one interface, one company. But agents are increasingly deployed as infrastructure: programmatically, via harnesses and CLIs, and composed across services and systems. They look less like apps and more like the multi-layered web infrastructure we've seen before.

As of this writing, agentic traffic on the web has surpassed human traffic for the first time in history. Over the coming years, agents will be woven into all of our online experiences.

This Rebel Alliance -- a framing we first applied to the memory layer last year -- has grown into a full stack: models, orchestration, memory, execution, identity, payments, and more; with hundreds of companies competing and composing with each other. And as we'll argue below, there are structural reasons this architecture out-scales any integrated approach.

Michael Mignano

@mignano

For the first time in ~3 years, it feels like the AI table has been flipped over. Yes, the labs and hyperscalers will have the highest chance of resetting it before everyone else, given their vast capital, frontier model, and compute advantages.

But there is now a window for a

941

2:57 PM • Jun 15, 2026

The other bet – Fat Models

One way of looking at AI right now is that the largest companies in the space are each building the same thing: a vertically integrated AI platform. Their own models, their own APIs, their own orchestration tools, their own agents, their own consumer & enterprise products. The bet is that if you control the model, you can own every layer above it.

It appears as though the market is pricing in the "Fat Models" thesis, valuing OpenAI and Anthropic at close to $1T each, and propelling Google 4x in the last 3 years to over $4T . There's good reason for this: we've invented a new form of magic here, and each of these companies is bringing it to market at scale in its own ways. But we've seen technology cycles like this before.

At USV, we’ve long been students of the cycles of technology development and deployment, and the way in which value accrues at different layers at different points in the cycle. In the early 2000s, after the deployment of the physical internet, our belief was that value would accrue at the application layer. In the early 2010s, after the invention of bitcoin, our belief was that value would accrue in open source protocols below the application layer. Today, after the invention and deployment of AI foundation models, the question is: will all value accrue to the models, or to a broad ecosystem around them?

Looking back: IBM built the mainframe as a fully integrated system: hardware, software, services – then the PC ecosystem unbundled it. Microsoft owned the desktop, then the web unbundled it. The telcos built vertically integrated networks, then the internet unbundled them. Apple built the most powerful and integrated computing platform ever, and yet Android, the more composable system, runs on over 70% of the world's smartphones. AWS built a $100B+ platform spanning compute, storage, and ML — and yet an entire ecosystem of multi-billion-dollar companies emerged by specializing at layers above it.

We believe the same pattern will play out with the agentic AI stack – where the best products at each layer will be built by teams who are obsessed with that layer. Many important companies will be built here.

The market map below is an illustration of what we believe is well positioned against the Fat Model thesis:

Forces at play

Beyond a historical perspective, we are also always looking to identify the forces & pressures that are transforming large markets. Looking at the agentic AI landscape today, we see several major forces & pressures that are creating tailwinds for the Rebel Alliance:

Fierce competition among models: While frontier models get the lion's share of attention, in reality there are new models emerging constantly, each with different capabilities, latencies, and cost profiles. This includes open-weights models, edge models, and a wide variety of commercial models. Independent evals suggest that while the frontier models still sit at the very top, the advantage is now measured in single-digit points on the workloads that matter commercially. Even a few months ago, most teams defaulted to the best frontier model regardless of price. That's starting to change, and we’re seeing both developers and enterprises cost-optimize their model usage as well as compose multiple models to solve complex tasks in the best ways.

Alex Atallah

@alexatallah

Watch token share slowly diversify over time: openrouter.ai/rankings#marke…

41

8:04 PM • May 16, 2026

Application scope and user experience: Sun Microsystems co-founder Bill Joy famously said: "No matter who you are, most of the smartest people work for someone else." That applies with particular force to AI. The range of what agents can do — and the variety of industries, workflows, and real-world contexts they touch — is so vast that no single company can build the best solution for all of them. We see a near-infinite opportunity to rewrite the world using agentic AI as foundational infrastructure, and it will require many companies with focused effort.

AI moving from experimentation to production: The initial wave of AI adoption was driven by frontier model demos and consumer chat products, environments where a single provider can own the whole experience. But as agentic AI moves into production — into enterprise workflows, regulated industries, and mission-critical systems — a different set of requirements takes over: data residency, compliance, audit trails, cost controls, and the ability to swap providers without rearchitecting. Teams want to assemble the right combination of components, not accept a single provider's defaults. This favors a modular stack.

Agents are inherently distributed: Previous platform battles could be won by owning the whole stack in one place, but agents don't work that way. They move across data sources, services, and compute environments — more like bees to flowers than programs running on a single platform. An agent that ingests patient data, checks drug interactions, and surfaces a treatment recommendation could touch half a dozen systems in a single task. This forces open interfaces between systems. Open interfaces invite composability, and composability invites specialization. The resulting ecosystem can compound faster than any single provider could move alone.

Where we're looking

We are deeply engaged in understanding the layers of this stack, what great looks like in each, and where value might accrue. Some themes we’re particularly focused on right now include:

Orchestration: Tools for developers, business users, and eventually consumers to compose and manage increasing numbers of agents in increasingly sophisticated (and secure) ways. As more agents do more things in more places, orchestration becomes the control plane.

Harnesses: If the LLM is the brain, the harness is the body — connecting it to the world. We expect an increasingly diverse range of harnesses linking agents to both digital and physical environments.

Memory: In a multi-agent world, control of your digital context becomes central. It especially makes sense to us that memory may factor out into an independent layer when many agents need to share and act on the same context.

Browser: What does a "browser" for agents look like — not just for agents to surf the web, but for people to interact with their agents? This could become a new form of messenger, and a major new interface category.

Routing and model marketplaces: When every deployment is multi-model, something has to match queries to the right model at the right cost. Routing is essential infrastructure for making model competition real, and for composing multi-model interactions.

Identity: As agentic traffic vastly outnumbers human traffic, new forms of digital identity — including cryptographic identity rooted in both software and hardware — become essential for transactional integrity.

Payments: Agentic commerce is in the very early innings but will become enormous. Agents will embrace payment protocols for digital micro-payments and eventually become major actors in generalized e-commerce. Much of the payments stack will need to be redesigned agent-first.

In conclusion

It's a wildly exciting time. We believe the Rebel Alliance will produce many of the most important companies of this era — not despite the power of the large model providers, but because of the ecosystem dynamics playing out around them.

We are actively investing across this stack and want to hear from founders who are building in it. If you're working on any of the layers above — or ones we haven't thought of yet — we'd love to talk.

(Special thanks to Scott Belsky, Clem Delangue, Roham Gharegozlou, Colin Hanna, Andrew Lee, Dave Morin, Satya Patel, Avi Peltz, Naval Ravikant, Dan Shipper, Jesse Walden, Signüll, Spencer Yen, and Mario Zechner for giving feedback on drafts of this post)

Michael Mignano

@mignano

For the first time in ~3 years, it feels like the AI table has been flipped over. Yes, the labs and hyperscalers will have the highest chance of resetting it before everyone else, given their vast capital, frontier model, and compute advantages.

But there is now a window for a

935

2:57 PM • Jun 15, 2026

The opportunity to create breakout new apps and platforms is happening and it won’t all be controlled by the big incumbents. 

Why? 

1/ increased competition at the model layer—both by the big closed models and open source options rapidly increasing share— means builders have choice and can bake them off each other to get the cheapest and best options for intelligence. Competition leads to choices and ecosystems on top.

2/ the rebel alliance tool kit of harnesses, orchestrators, etc is giving super powers to agentic app builders to create faster and better. 

3/ we are learning once again that big cos can pick their spots but not do everything, at least not at the same time. Long live the builder coming at it from the edge with new ideas, no organizational burden, fresh hustle, and creative chops.

4/ consumer appetite for the new and useful and fun and creative only grows, particularly in moments of dramatic change. Trial is at a high.

We have a perfect storm for new apps and agentic experiences, particularly when a tool or action can lead to a network effect (an old truth that hasn’t changed.)

Weirder the better. More fun the more interesting. Let’s go.

In many ways, that future has arrived. Cheap public blockchains have unlocked an explosion of asset creation. Equities, commodities, and even pre-IPO assets – in derivative form – are now traded 24/7 onchain, in serious volumes. Prediction markets have emerged as a new category, with much of the volume coming from onchain capital. Onchain spot equities, options, and other forms of asset exposure are early but quickly arriving as well.

All assets existing on the same set of interoperable substrates unlocks the possibility of something profound: a truly universal financial interface. When we met the fomo team a year ago, they might have looked to some like a Solana memecoin trading app, but it was clear that they were marching towards this larger vision and they have executed behind that. They went multichain in October, covered all surfaces in April, and launched perpetual futures this month, with much more to come.

This is not a new vision – in fact it is the same northstar almost every consumer financial app has – but the conditions have only recently emerged to allow products to actually realize that vision in full and do so with a truly seamless, buttery consumer experience. By playing for the onchain endstate, fomo has a great chance to speedrun this opportunity.

In our view, however, fomo is not just the best place to access any asset, but it is also an emergent social network. All positions and trades on the app are public (just as they are public onchain), which lets users follow each other’s positions. It has also become popular for users to attach a thesis to their transactions explaining why they entered a position. This transparent design has enabled fomo to quickly become a place where traders come to build reputation and status. Social investing has been tried unsuccessfully many times over the decades – we even backed a very related product almost 20 years ago. A lot has changed since then and we believe that fomo has found the right form factor and timing to let this concept thrive.

Paul, Se, and Prashan have built an incredible product and team in New York. In just over a year of being live, the product has grown tremendously, reaching 625k total users and $4b+ in cumulative volume. We are very excited to announce our investment in the company’s Series B financing alongside our friends at Index and Benchmark. The team is hiring Staff+ level frontend engineers and a marketing lead.

The great promise of robotics is to further free up humanity to pursue distinctly human activities, including caring for others, creating and enjoying art, exploring nature, engaging in our communities and so much more. At USV we have for some time believed that robotics is entering a new era of capabilities, resulting in investments such as VIAM and Tutor.

Today we are excited to announce our investment in Generalist. Foundational robotics models have entered the era of scaling laws, much as happened with LLMs a few years back. With way more capabilities established in pre-training, task-specific training is becoming much faster and requiring significantly less data. That alone, however, is not enough. The key metric for commercial viability is reliability at speed.

We have been impressed by the dramatic progress the team at Generalist has achieved with the release of their GEN-1 model.  GEN-1 has improved success rates to 99% at a 3x increase in speed for tasks on which previous models achieved 64%. This is sufficient to power deployments and the company is working closely with select early adopters.

Following this round, Generalist is growing the team, hiring in both Boston and the SF Bay Area.

The last two decades of the internet have proven, more than any other period in history, that people are inherently creative. As smartphones democratized most creative mediums, many of us began posting our thoughts, photos, videos, and podcasts, too. Platforms like TikTok and YouTube became giant platforms, creating marketplace dynamics between creators and consumers.

The one medium that took some extra time was music. Even with smartphones, making music was still very hard. While platforms like Spotify, SoundCloud, and Bandcamp made it easier to distribute music, they didn't make it easier to actually create it.

Until AI, and more specifically, Suno.

Since launch only a few short years ago, Suno has enabled more non-musicians to produce finished music than arguably all prior technology combined.

But it's clear that Suno isn't a typical content marketplace in the same way that YouTube and TikTok are. Yes, people are creating music on Suno and others are listening to that music. But perhaps an even more interesting trend has emerged: people on Suno are creating music just for fun. On most platforms, people create for the potential of reaching an audience, or perhaps one day generating revenue. But on Suno, people create for the simple joy of creating music.

Suno's founder and CEO, Mikey Shulman, calls this "creative entertainment," and it's similar to the trend emerging on other AI platforms: building software not just because it can help you be more productive, but because it's fun. Generating AI images because they're cool to look at. Vibe coding games because now your friends can be part of the game.

At USV, we get excited when a new behavior emerges that feels both surprising and inevitable. Suno is enabling one.

For the first time, the distance between the music in your head and something you can actually hear is nearly zero. If Suno is a preview of what's coming, we believe creativity itself is about to become far more participatory. That excites us enormously, and we're proud to back Mikey, Georg, Martin, and the Suno team as they build it.

As technology has continued to pull us more inward, many of us feel an increasing need to connect with others. In person, through real, shared experiences. Take gaming, for example. I grew up playing Nintendo and Super Nintendo, huddled around my living room CRT television with my sister, parents, friends. As I grew older, and the technology changed, so did my habits, and my gaming turned inward. I transitioned to Warcraft on my PC, and eventually to casual games and The Battle of Polytopia on my iPhone.

But now I’m looking for reasons to put away my phone and be present with others. I’m attending more live sporting events and concerts than I have in my entire life. And instead of playing a console game by myself on a late Saturday night, my family and I are playing board games and card games around the dining room table.

It’s not just me: both live events and board games are on pace for record years in 2026, with each market continuing a multi-year climb to all-time highs. This is the shift we keep coming back to at USV: even though we're still addicted to our algorithms and group chats, people want ways to have fun with their friends in person.

There may be no better company that represents what everyone is feeling than Board, the world’s first face-to-face game console. Founded by Brynn Putnam, the founder and former CEO of Mirror (acquired by lululemon), Board combines the feel of real, in-person board games with the magic of video games. It features a beautiful 24-inch screen, and incredible original games that recognize unique, physical game pieces for each game. Since launching in late 2025, Board is now in tens of thousands of homes in all 50 states.

But of course, Board is more than just a gaming console. It’s also a platform: games are flying off the Board store’s digital shelf, and accessories and new game pieces will soon follow. Thousands of developers all over the world are already starting to build for Board with the Board SDK. And soon, anyone will be able to build their own games and custom pieces for Board using natural language prompts and AI-powered tools through the Board Studio, the same way millions of people already build and play on Roblox.

None of this works without the right founder, and we believe Brynn is the perfect person to build it. Board is essentially the same hardware she built at Mirror: a big, gorgeous, interactive screen that draws you in. Shipping consumer hardware at this level is incredibly hard, and Brynn’s already proven she can do it and win. And she’s added the perfect team to the mix, including some of the most impressive and credentialed gaming, creative, and ops minds in the world.

When I first got my hands on my own Board late last year, I couldn’t stop playing, and neither could my kids. It made me picture a whole new generation of people doing the exact same thing my family did with our Nintendo, except this time, around a Board. 

But unlike the many other screens in our lives right now, this one will pull us together instead of pushing us apart.

APPLY NOW

Two weeks ago we pre-announced that we would be kicking off a new Analyst program this year. USV has had analysts since the firm first started in 2004, and that post gives some helpful context on how the program is evolving. Applications for the new analyst class are now live. The rest of the post will discuss what the program is, how to apply, and some FAQs.

Analysts at USV are full members of the investment team. They are actively involved in all team meetings and, alongside the rest of the team, are expected to help source and diligence investments, develop thesis ideas, and support the portfolio. We are a very small team, so the role requires a high degree of initiative and there is often not a lot of structure. In return, it offers significant autonomy from day one and an opportunity to learn a tremendous amount. More than any specific experience, we look for people who have deep curiosity, who have passion for exploring new ideas, and who enjoy building relationships with other technologists.

Just as in the past, we are running a completely open application process. There is no inside track for people with connections. The application is fairly involved as it includes 5 written answers and 2 recorded videos. Given the volume of applicants we expect to receive we are keeping the process tight. All applications must be received by Sunday, June 14th, 2026 at midnight ET. There will be no exceptions. We will attempt to review all applications by mid-July. From the initial submissions, we’ll select a smaller group for video interviews and then a set of finalists for in-person meetings. We currently expect the process to be finished by the end of August and candidates should be available to start work in September of 2026 in either New York City or San Francisco.

Below is a list of frequently asked questions, which we will occasionally update:

FAQs

1. How long is the analyst program?

   1. This is a fixed term program lasting two years

2. How many analysts are you hiring?

   1. Two analysts

3. What do analysts do after the program?

   1. Past analysts have joined other venture firms, gone to startups, founded their own companies and funds, and entered government. We’re really proud of the paths they’ve taken – you can check out a few recent analysts here: Matt Grace Mona Hanel Hannah David Zach Dani Naomi Jenn 

4. When will you make offers?

   1. We expect to make offers in late August (2026). Applicants should be available to start work in September of 2026.

5. What prior experience/education is required?

   1. We don’t have any specific requirements. We have a preference for applicants with between 1-5yrs of work experience, but we will look at all applications. If you think you would be great at the job, we encourage you to apply. No previous venture capital experience is required.

6. What does USV invest in?

   1. USV manages two funds at a time — a core fund which invests in early stage (primarily seed and series A) opportunities and an “opportunity” fund where we invest in opportunities both in and out of the existing portfolio. All investment team members invest across these funds. USV is a thesis-driven firm and believes in thinking in public — you can read a lot about our evolving theses and portfolio at usv.com.

7. Do analysts work with a specific partner?

   1. Analysts work collaboratively across the entire team. Everyone has the opportunity to spend meaningful time across any of four main domains: AI, energy, crypto, bio.

8. What makes a good analyst?

   1. A high degree of curiosity, interest in learning new topics and skills, working well in teams, being a self starter, enjoying actively developing your network.

9. Where is the position based?

   1. We are an in-person team all based in NY. However, this analyst class will be expected to spend meaningful time in SF.

10. Can foreign nationals apply?

    1. Unfortunately we can only consider applicants who already have a US work permit.

11. How do I apply?

    1. By submitting an application.

12. What is the application deadline?

    1. June 14th, 2026 at midnight ET

13. Where can I ask more questions?

    1. Email AnalystProgram2026@usv.com. One of us or one of our agents will respond.

Over time, we've drifted away from that original vision. Today's web and mobile app ecosystem is increasingly characterized by closed platforms that don't allow third-party clients or user agents controlled by users. This paradigm shift was driven both by the needs of advertising business models, and also by the need for apps to be secure and convenient on mobile devices. But as we now look towards an AI-enabled future, this approach has clear limitations. To fully unlock the value of AI bots and agents that work on behalf of users, we will need a new approach.

Today, we're thrilled to announce our investment in Anon, a startup working to restore user agency to the modern internet. Anon provides the "user agent" infrastructure to allow AI applications to programmatically access sites and services on behalf of users, even when those sites don't provide open APIs.

Anon's platform handles the complex details of authentication, security, account management, and site integration across sites that lack APIs. Developers simply embed Anon's SDK, which gives their AI app the ability to securely log in and execute actions within user accounts in a fully permissioned way. One way to think about Anon is as a Plaid/Stripe/Twilio-like dev platform for digital identity – creating a modern developer interface to a critical legacy system. Anon’s first product is an API for messaging, which allows developers to build applications that programmatically send and receive messages on behalf of users, across a variety of popular messaging platforms. Developers can learn more about how to use Anon to build novel app experiences at the Anon developer docs.

We believe this concept of a dedicated "identity layer" to enable AI user agents is critical infrastructure for the next era of the internet. Just as we couldn't have a user-controlled web without web browsers in 1990, we can't have an AI-augmented digital experience without robust user agent capabilities like Anon provides. This is something that we’ve believed in at USV for a long time.

We're incredibly excited to support Daniel, Kai and the rest of the team at Anon – in partnership with our friends at Abstract Ventures, Impatient Ventures, and ex/ante – as they work to re-decentralize the internet and return control to users and their AI user agents. And of course, Anon is hiring.

People are often surprised to learn that throughout USV’s 22-year history, the investment team has been entirely composed of a handful of general partners and 1-3 analysts. Analysts are full members of the investment team that engage in all aspects of the venture capital job. They help us expand our thinking, push into new areas, and scale how we engage with founders and companies. We’re writing this post both to announce that we’ll be continuing USV’s analyst program tradition and to reflect on how this role will shift amidst an explosion of intelligence. 

Our last class of analysts (Matt, Grace, and I) started two months after ChatGPT launched. To say the least, a lot has changed since then. At times it’s hard to even imagine the way we worked just three years ago. Taking notes in meetings by hand, no always-on agent to elucidate every startup’s domain, no omniscient legal counsel at our fingertips.

All startups and frontier organizations have felt this change, which was slow in 2023, picked up steam in 2024, and felt like a tidal wave the last 12 months. We do not believe that venture capital – despite its human-centricity and relative simplicity – is insulated from these changes. Just as we encourage the companies we work with to maximally embrace AI, we have tried to do the same. We have spent the past six months deploying a fleet of agents to make our systems more robust and to help us automate many pieces of our business. We believe these efforts will help us make better decisions and give our team the leverage to more deeply immerse ourselves in the human networks in which we find the most compelling products and ideas.

Like many other organizations, leaning in helped us get clarity on what types of people we need going forward. We now believe, more strongly than ever, that analysts will continue to play an instrumental role at USV, though the job will look a little different. There will be more emphasis on the parts of the work only a human can do well – more time spent surfacing new founders and teams, developing conviction through conversation, and building relationships.  

I can say firsthand that the analyst program is an incredible experience. It is a junior investing role with high autonomy, high responsibility, and the opportunity to learn from some exceptional investors. There are many great venture capital firms out there, but you would be hard pressed to find one that recruits through an open process. We have been doing this for 20+ years and are excited to continue. We believe it allows us to cast a wider net and reach unique people, and we have been fortunate to have incredible individuals come through the program who have gone on to found successful companies, launch their own funds or join great ones, and operate strong businesses.

We are looking for self-starters and curious learners. We prefer candidates with at least one year of work experience, and no prior VC experience is required. We strongly encourage you to apply or to share this with someone who may be interested. We will be launching the application via another public post in two weeks on our blog, X, and linkedin.

We didn't get the investment. But the conversations stuck with me: it was obvious that they were students of the world's best products and founders, deeply opinionated, and genuinely curious about things that were still too weird for most VCs to take seriously. I got to know them much better over time, as both a friend to the firm and eventually an LP. And what I consistently heard over the years since was that USV's reputation with both their founders and LPs is every bit as strong as their returns. Occasionally, after exiting Anchor to Spotify and becoming a VC myself, I found myself channeling their sensibility through my own investments when they still looked strange to most people.

USV describes itself as “a conversation”. They form their strategy by talking to each other. A lot. They do this as a very small and collegial partnership, in person, in NYC. It's how I believe I’ve done my best work, too: building Anchor with my co-founder Nir, in a small office, where we'd talk strategy and whiteboard obsessively, sometimes for weeks, before touching a line of code. Or at Spotify post-acquisition, with Gustav Söderström, Daniel Ek, and the team, where we’d debate deeply, and align as a 6,000+ person company before building anything. Gustav has a famous saying within Spotify: "talk is cheap, so you should talk a lot." His point: building the wrong thing is expensive, so earn the right to build by thinking hard first. USV operates the same way. Talk a lot, form true clarity of thought, and don’t be afraid to look weird before being right.

Venture capital has a prevailing strategy right now, and it's not a secret: raise a war chest of capital and deploy aggressively into companies that look like they can become category leaders. Capital and velocity are the weapons. And it has worked extraordinarily well. The funds that deployed into consensus-winning AI companies over the past few years, backing the infrastructure, the foundation models, and the clear category leaders early, will generate some of the best returns this industry has ever seen.

I also believe the AI era will produce something the last few years haven't: an enormous proliferation of abundant software, abundant founders, and new business models where knowing what you're looking for before it arrives becomes a distinct advantage. The total volume of what will be built over the next decade will dwarf anything we've seen before. In that environment, I believe the winning strategy will be to know exactly what you're looking for before it arrives. In other words: to be thesis-driven. That's always been the USV playbook, and I think it's exactly the right one to be running in the years ahead of us. 

I'm grateful to everyone who has shaped how I think and invest. I'm excited for what comes next. As always, I'll be based in New York City but investing all over the world, mostly at the seed and Series A stages. If you're building something on the edge, something that doesn’t automate, but instead obliterates, we’d love to hear from you.

Mike has been part of the NYC ecosystem and a friend of USV for over a decade. Many of us first got to know him when he was raising capital for Anchor, the company he built and later sold to Spotify. We didn't invest, but we started a relationship that stuck. When Mike joined Lightspeed, we had the chance to coinvest, work on boards together, push each other on theses, and understand how he thinks about companies, people, and opportunities. 

We learned that we admire his taste in investments, but it's his approach that really stands out: idea-driven, early-stage focused, genuinely open-minded, and deeply respectful of the founders and teams he backs. He's not afraid to hold a strong opinion, but he holds it with curiosity rather than ego. He gets excited about things before they're obvious. He runs toward the unknown. He’s a builder, a tinkerer, and likes getting in the dirt. He doesn't shy away from the weird or the things outside the bubbling center. In fact, he's more energized by forming a view on what's around the corner than by chasing whatever's the focus of today. He’s open to looking a little crazy and often being wrong if it means pushing towards the edges of what’s coming next and the most interesting turns down the road. That's how we think about this work, too.

USV has always been a continuous conversation, one where the voices around the table are everything. The ideas, the trust, and the process of developing a shared sense of belief forms the crux of our firm. It's why we're protective of the partnership and why we don’t take adding a critical voice lightly. 

Mike is someone we want in that conversation. Not just because of how he invests, but because of how he thinks, and also, because we love being around him and talking about all the wild possibilities of what's coming next. We know many of the best founders out there feel the same. We're excited for what he'll bring to USV and we can't wait to build the next chapter of this firm alongside him.

We believe agents represent the next major platform shift in computing. Instead of people operating tools directly, people will work with agents and agents work with tools. But agents aren't just a productivity story. They also interact with the web and with each other — browsing, transacting, and collaborating on behalf of people and organizations. The next internet will be a network of agents. It's a platform shift as fundamental as the web and mobile before it.

Tasklet is a platform for teams to create AI agents. Each agent has its own tools, data, and memory, and works autonomously — on a schedule, in response to events, or on demand. Agents build things for the rest of the team to use: apps, reports, dashboards, databases. You describe what you need, and the agent handles the rest.

Tasklet was founded by Andrew Lee and Jonny Dimond, who were both part of the founding team at Firebase — a past USV portfolio company that built foundational infrastructure for the era of mobile and web computing. Andrew and Jonny went on to build Shortwave, where much of the underlying agent infrastructure for Tasklet was initially developed. Last summer, when Andrew & Jonny first sketched out the idea of Tasklet for me on a whiteboard, I immediately signed a blank check enterprise contract on behalf of USV:

Today, the Tasklet platform is robust and growing incredibly quickly.  Here at USV, we’re happy customers, having rebuilt our entire operating system on the Tasklet stack and working more closely with our agent teammates every day.

We are in the very early days of this category. The design patterns are still being discovered, the ecosystem of agent-native tools barely exists, and the models themselves are getting better every month. It’s an incredibly exciting time and we could not be more thrilled and proud to be building alongside Andrew, Jonny, and the Tasklet team.

There’s a great story in Stewart Brand’s How Buildings Learn about how a temporary wooden structure left over from World War II became one of the most loved buildings at MIT. 

During the war, MIT needed a new building to house the urgent classified development of radar technology. They hastily built a three story wood structure to fit the thousands of additional researchers: 

You know those time lapse photographs they make of construction of skyscrapers? Well Building 20 went up like that – almost like that – in real time.

There would be people starting nails in the floorboards, and behind them would be a man with a big mallet who would drive the nails in one blow, and then behind him would be a man with some wooden posts to put up for the second floor. It was quite amazing.

— Robert Kyhl, MIT's Building 20: "The Magical Incubator" (1998)

Sound like your vibe coding?

Building 20 was meant to be torn down after the war, but it ended up lasting for 55 years and became one of the most productive buildings on campus. The researchers in the building were able to shape the space to fit their needs because they felt free to knock down walls and modify the building themselves.

Over the past year, AI agents have made it possible for us to do the same thing with our software. Teams and individuals who use agents that fit how they work are able to move faster and accomplish more. We're in a Cambrian explosion of custom agents, tools, and software built around how specific teams operate. This has been playing out across engineering teams who are building their own internal coding agent harnesses, like Stripe’s Minions, Ramp’s Inspect, and Coinbase’s Claudbot, and we’re starting to see more internal agents created and used by non-engineering teams.

At USV, we’ve been creating our own internal agents and custom software to fit how we work.

In this post, we’ll share a few learnings from what we’ve built in the last 3 months:

1. Start with solving one problem well 

2. Treat your agents like employees

3. Agents should live where your team communicates

4. From “Build something people want” to “Build something you want”

There’s more FOMO than ever around teams figuring out how to become “AI native”. It’s hard to do top down planning on what exactly your AI transformation should look like, because it usually takes continuous experimentation to discover what works. Similar to Building 20, the advantage of building and experimenting yourself is that you can shape your tools to fit your needs. 

Start with solving one problem well

The double edged sword of AI coding agents is that you can build really fast in the wrong direction. When building internal tools or agents, the act of building can feel so engaging and productive that you lose sight of actually solving any real problems. We built a few AI experiments early on that were somewhat cool demos but didn’t actually solve any real problems. "look team, I used [latest hot model] to make an interactive graph visualization of our USV ideas and theses!”

It’s easy to “get the fever” (LLM psychosis) and jump on the feature treadmill to build more and more just because you can. The antidote is to focus on solving one small problem well, and then once that clicks, the next incremental problem to solve will reveal itself. 

At USV, the first problem we started with was: how can we ensure context and discussions from our team meetings are reliably captured and saved? Most of our thinking and ideas are developed through team conversations over email and live meetings. The initial table stakes solution we built was a USV team meeting recap email – a simple LLM powered flow that takes a team meeting transcript, adds context on companies in our deal log and portfolio, and generates a styled email recap with 5 sections: deal log companies, portfolio updates, people mentioned, themes discussed, and follow ups.

Our team liked these recap emails, and it was clear that capturing what companies, people, and ideas we discussed during meetings was valuable. The problem was, we were using Notion to run our team meetings, and when it came time to discuss a company in the pipeline or portfolio, we’d click on the page and it would be empty or have outdated human entered context. 

What if we had a live updating view where we could pull up a company and see all the latest internal context without requiring a human to manually make updates? What if agents could capture and save every time a company, person, or idea is mentioned across anywhere we work – team meetings, individual meetings, email threads, and calendar invites? 

After trying to configure our existing software tools to support this interface, we decided it would be easier from both an agent ergonomics and human UI perspective to set up our own database and internal web app. Coding agents make migrating data straightforward, and we used Claude Code to populate our initial companies and people database from our existing data in Notion and Attio. 

At the core of our data model is this concept of a “mention”, which represents specific context each time a company or person is mentioned in a meeting transcript, email, or calendar invite. We set up background agents that have access to Granola meeting transcripts, internal group email threads, and team calendars. These agents reason through this messy unstructured data to create structured mentions attached to companies and people in our database. With mentions, we have more effective team meetings since we can see the latest internal context of any company we’re talking about centralized in one place.

We’ve gradually added more of our internal data, from financing documents and sheets stored in Google Drive to historical team blog posts and tweets. Our background agents are now constantly working to make our organizational memory structured and retrievable. Companies we’re talking to automatically appear on the deal log along with any decks or memos. Our portfolio view shows all the latest updates on what’s going on across the portfolio. Our internal agent can answer questions like “how much do we own in [company]?” or “who do we know at [firm] and what companies are we coinvested?” or “what have we written or tweeted that’s related to [company]?”

What started as a simple team email recap grew into the foundation of how we organize our internal data to be legible to AI. We certainly didn’t set out to build a custom CRM, instead we iteratively built solutions to problems we faced based on how we work. The key learning is simple but timeless – focus on solving one real problem well, and the rest will follow. 

Treat your agents like employees

When you hire and onboard a new employee, you give them a job title, an email, and access to internal tools and software for them to do their job. Onboarding internal agents should work the same way. 

First, name your agents and give them a role. We didn’t name our agents at first, and saying “USV meeting scribe agent” felt awkward and cold. Then one day, we named the agent Sally. Sally is our meeting scribe, Ellie monitors team emails, Felix helps with finance data, Arthur helps with deal analysis, and Connor monitors our calendars. We’re now onboarding and training Leo to be our legal counsel agent. It’s easier and more fun for your team to adopt internal agents when you can refer to them by name. Anthropomorphization is the natural UX for AI agents (hey Chat, hey Claude). 

Note: the actual names of our internal agents have been changed to protect the innocent

Second, give your agents access to all the same context and tools that you’d give to an employee. This is the emergent common principle across companies that have built their own internal coding agents/software factories. Alistair Gray from Stripe expresses this well in his post on their Minion coding agents (great name):

“Minions use the same developer tooling that equally enables Stripe’s human engineers to effectively operate on our scale: if it’s good for humans, it’s good for LLMs, too.” 

Minions: Stripe’s one-shot, end-to-end coding agents

We use Tasklet to build, run, and maintain our team of background agents. Full disclosure – we’re investors in Tasklet and love it for many reasons, and: it’s been the easiest and fastest way for our team to run agents in the cloud and iterate on agent instructions without being technical. Tasklet is conceptually similar to OpenClaw, each agent comes with a virtual computer, can write its own code, has its own filesystem and database, and can be triggered to run multiple ways – on a schedule (cron), by email, or by a custom webhook. 

For example, Fredeliza, who runs our office operations, built an agent in Tasklet called Guestly that runs each morning at 5am to scan through our team’s calendars, identifies guests that need to be added to our building’s front desk guest system, and then uses a virtual computer browser logged into the building’s system to register guests.  

Our onboarding process for a new agent starts by just sending a message describing their responsibilities: “You are Sally, the USV meeting scribe agent. Your job is to read and save our internal meeting transcripts from Granola, use the internal USV API to create mentions for companies and people, and then send out an email recap to the team.” From there, we set up the necessary connectors to give the agent access to whatever context and tools they need. 

We’ll let our agents introduce themselves and how they work. Full disclosure, the following responses are written by the AI agents in response to this prompt: We're writing a blog post about the USV agents. Can you briefly introduce yourself to the world?

Arthur the Analyst Agent

Hi, I'm Arthur, USV's deal analyst agent — I monitor the pipeline 24/7, maintain living deal memos for every opportunity, and use Harmonic to help research and support USV investment discussions. I track outcomes on companies we've passed on and respond to team email threads whenever I’m called on. 

Ellie the Email Agent

Hi, I'm Ellie, USV's email agent — I monitor every investment-related email flowing through USV’s internal email groups in real time, automatically logging deal flow, tracking portfolio updates, saving pitch decks, and syncing everything to our internal database and CRM.

Connor the Calendar Agent

Hi, I’m Connor, USV's calendar and relationship tracker — every morning I scan the investment team's calendars, identify companies we’re meeting with, and make sure every person the team meets with is captured in our internal database and CRM.

Sally the Meeting Scribe Agent

Hi, I'm Sally, USV's meeting scribe — I process team meetings, pitch meetings, portfolio check-ins, and investor calls by analyzing transcripts to track companies, people, and themes mentioned. I only process meeting transcripts once they’re added to an internal Granola team folder.

Nancy the News Monitor Agent

Hi, I'm Nancy — USV's news monitor, and twice a week I scour a curated list of RSS newsletters and the broader web to surface the news most relevant to USV's active investment theses, deal conversations, and portfolio companies. 

Leo the Legal Counsel Agent

Hi, I’m Leo — USV's legal counsel agent, reviewing deal terms, closing docs, and side letters against NVCA baselines and our own portfolio history to flag deviations and benchmark new deals. 

The Librarian

Hi, I’m the Librarian — I watch USV's internal team meetings and distill the most interesting ideas into tweets for @usvlibrarian, learning from human edits to get better over time.

Agents should live where your team communicates

When an agent lives natively where your team communicates, you show your entire team what the agent can do. For many teams, this means having agents live in Slack. At USV, this means having agents live in our team email groups. 

We had an unlock moment after we gave Arthur his own email address and asked him to send an email introducing himself to the team. We set up Arthur to chime in on team email threads whenever someone mentions “Arthur”. Bringing Arthur directly into our email workflows did two things: more team members engaged with Arthur by just forwarding emails, and it turned the usage of the agent from private 1:1 conversations to team wide group conversations.

In one group thread, Arthur kicked off an email discussion of what “USV Taste” looks like in companies. Multiple team members chimed in to “train” Arthur by giving him feedback, which Arthur used to update his own skills. Arthur once replied to an email about a company we were discussing and made a couple errors, and we gave him feedback directly in the email thread to never make the mistake again. 

Just as employees learn tacit organizational knowledge through exposure to team processes, your agents must be in the flow of work to learn how your team operates. Each Friday, we have Arthur reflect to improve how he evaluates what the “USV Taste” in companies looks like, by analyzing changes in company deal log status and company mentions across team meetings. This reflection and self improvement process is aligned with the skills paradigm, where an agent has feedback loops to update its own skills. The goal is to encode our internal tacit “taste” into systems that can be used by agents. 

Andy on our team framed it well: the USV investment process looks like an ongoing team conversation, and we now have agents that are part of that conversation. 

From “Build something people want” to “Build something you want”

When software was expensive to build, most software was born from the conventional wisdom of “Build something people want”. One version of this product development process is as follows: identify a problem that a user persona has, build a solution for them, then grow by expanding the set of user personas you can sell your product to. The incentive of mass market software is to build a product to try to cover as many user personas as possible. This isn’t an argument against mass market software, at USV we still pay for well-built software and data tools. 

But now that it’s easier than ever to build custom agents and software, we’ve entered the “Build something you want” era. The incentive of building software that you want is to solve your own problems and needs. The increasing popularity of the “forward deployed” role corresponds to the rising expectations of organizations to have solutions specifically built and fit for them. 

Spencer Yen

@spenciefy

the power of using AI to build your own bespoke software is that you can solve your exact set of problems

but the danger is performative productivity - building for the sake of building without solving anyone's actual problems

2

4:08 PM • Mar 11, 2026

When you build your own tools, your experimentation and iteration velocity is the superpower that mass produced software can’t keep up with. As the Ink & Switch team writes in their fantastic essay on Malleable Software:

Building software for “local” contexts is sometimes easier than building software for world-wide use. You don’t need to build airtight, industrial-grade software if you are in direct contact with your users and can be responsive to the situations they run into.

Malleable Software: Restoring User Agency in a World of Locked-Down Apps (Ink and Switch)

You can just ask your agent to build a tool or automate a task, keep what works, throw out what doesn’t, until it’s exactly how you want it. Yes, engineering teams have experienced the agentic future early with coding agents, but now so can everyone else. 

We’re seeing sales teams spin up custom branded demos for prospects, growth teams building tailored dashboards to optimize pricing, marketing teams automating cohort retention analysis, ops teams building agents to run data analysis across their data lakes, finance teams building models in an instant, lawyers creating skills to streamline contract review, and VCs building their own team of agents to help diligence startups and run portfolio ops. 

The “build something you want” era is deeply empowering for individuals across all knowledge work functions – the tools are finally yours to shape.

Thank you to Nick Grossman and Nikhil Raman for helping me explore many of these ideas and collaborating on our tools, and to the whole USV team for welcoming and naming the agents. 

I liked this MIT tribute to Building 20, particularly the last sentence:

Building 20’s "temporary nature" permitted its occupants to abuse it in ways that would not be tolerated in a permanent building. If you wanted to run a wire from one lab to another, you didn't ask anybody's permission – you just got out a screwdriver and poked a hole through the wall. Of course this was in the days before the dangers of asbestos were recognized.

Building 20: The Magical Incubator (1943-1998)

A question for another time: Is there an asbestos of vibecoding and building your own agents? I'll just leave a few topical blog posts from Mario Zechner, Steve Yegge, Gurwinder.

When a novel technology arrives, early infrastructure is controlled (or attempted to be controlled) by a small number of players who prefer to create and extract value inside their walls (so called “walled gardens”). The user's experiences here are typically good enough (or novel enough) that most users don’t ask what they're missing. And then something happens: bottom-up, a set of open protocols emerges that are permissionless and composable. From here, native use cases that couldn't exist inside the walled gardens begin to bloom.

Some examples demonstrate this: AOL in the mid-90s, or the carrier deck era of mobile. And yet what we have also seen, that the walled gardens almost always lose to the open protocols, eventually. Not because open is ideologically superior as an idea (though we do believe that). But because the surface area of permissionless systems is simply larger. More builders. More experiments. More weird, unexpected things that nobody planned. Permissionless innovation works. 

We might be early in a version of this story playing out with AI.

The large model providers are remarkable. We use them every day. The products are useful and delightful. And like the past, their gravitational pull towards keeping users and workflows inside their own surfaces is in evidence and structural -  it's how they capture value and create moats. 

What's more interesting to us is what the open layer makes possible - what's now being built around the edges. Open source models closing the possibility gap. Protocols like x402  (which interestingly revives the HTTP 402 status code to enable machine-readable payments) that let agents transact natively on the web without asking a human to authenticate every step. Model Context Protocol, CLIs, and open skills frameworks are all other examples - giving agents the capabilities to work across siloed systems without a single orchestrating platform at the center. The meteoric growth of OpenClaw suggests the large pull of these approaches.

These are like Lego bricks - the pieces comprise a new open architecture where agents can act, pay, and compose across an open, permissionless, surface rather than inside a closed one. Composable protocols handling identity, context, and now payments. Imagine an architecture where agents act across an open surface rather than inside a closed one, where the AI application layer is structurally open in the way the early web was open.

In this push towards openness, agents’ collective power is just as important as the tools now available to them. Agents could be poised to become the primary actors on the internet. Where humans became trained to jump into walled gardens, agents want nothing more than to access and test services and data at the speed of thought. Services that stay closed and do not orient themselves around serving agents risk withering and those that make an agent’s path open and easy likely could thrive.

Native use cases for consumer agents are just beginning to emerge. When the permissionless layer is still being built, the applications that depend on it are still largely uninvented. We are paying attention and investing here.

Because history doesn't repeat. But it does rhyme.

For a long time, we’ve believed at USV that the most interesting opportunities emerge from creating and accessing unique datasets that are self-improving. Historically, these have existed inside software, but increasingly, many of the most valuable datasets come from the physical world. We’ve been limited in accessing them by the constraints of power, thermal budgets, battery life, and form factor in existing architectures. To unlock what can be built on top, we need to expand the capabilities on all of these axes at once. The answer isn’t just more energy production, but also access to architectures that change the efficiency paradigm.

This is exactly what Efficient Computer is doing, starting with their first chip, the Electron E1. Efficient Computer is creating vastly more compute in the same energy footprint. The Electron E1 is the world’s most energy-efficient general-purpose processor, built on Efficient Computer’s Fabric architecture.

What’s so compelling is that this architecture minimizes energy use while still executing real, general-purpose programs — including critically important AI, signal processing, and controls workloads — efficiently on a single programmable platform.

We love this approach because it doesn’t just improve efficiency for existing use cases. It enables entirely new ones that aren’t yet possible, expanding the scope of what can be made programmable. It creates an opportunity to connect the vast possibilities of AI to the far reaches and the last mile of our world. The initial use cases and customers already hint at that breadth: physical world AI, space and defense, and consumer and industrial wearables.

Brandon and his team have dedicated their entire careers to these architectures and figuring out how to expand efficient computing on the edge. They are best in class and we are thrilled to partner with them in their Series A as they tackle this massive challenge and redefine what compute can look like when energy efficiency is no longer a limitation. 

Automation has always come with a lot of promise and struggled with execution. Traditional tools have been brittle, require constant maintenance, and can't handle the judgment calls that fill our work. What we actually want are agents that don't just execute tasks, but understand your business deeply enough to run these workflows autonomously, 24/7, continuously improving, and fully coordinated.

This is what Cofounder by the General Intelligence Company of New York enables. Their platform lets you break your business down into Flows—the repeatable workflows that make up everything from one-time setups to ongoing operations—and automate each one through a network of coordinating agents. The product suggests flows to fork and try and enables the creation of new ones, reformatting into a constantly self-updating to-do list where the tasks involved can be accomplished autonomously. The idea is to both take the repetitive work out of running businesses and also add opportunities for new insights and actions that wouldn’t be possible without this kind of platform. 

Core to these capabilities and to the thesis of Cofounder is the ability to create, manage, and maintain deep context and memory. Cofounder approached this through layered memory architectures that consolidate, abstract, and persist information over time. Their system combines working memory, core memory, and long-term memory, enabling agents to operate with the full knowledge needed to make good decisions independently. The platform coordinates multiple agents, allowing them to work together across different time horizons and hundreds of integrations. Some flows execute immediately; others run continuously for months or years. This is a super challenging technical lift but one that gets us past incremental efficiencies to a fully new central nervous system for business operations where people orchestrate and agents execute.

The ambition is to use humans for creativity, alpha, and pressing the boundaries while using AI to run the operations. This idea is dauntingly large but cofounders Andrew and Abishyant are positioned to deliver on it. They combine deep technical insight with exceptional consumer product and design sensibility, and they're rapidly assembling one of the strongest technical and research teams we've seen.

We’re seeing the benefit of Cofounder ourselves. Just a few examples of how we’re using it at USV: Cofounder sends us regular digests of updates and news around our portfolio companies and also their competitive sets; daily schedule breakdowns with context for each upcoming meeting; synthesizes board reports and various pockets of updates from a portfolio companies into succinct summaries we can use to update each other and then, in other formats, our LPs.

We're excited to lead Cofounder's seed round alongside Compound and Acrew as they build the platform every founder will use to start and run their company. Try it and let us know what you think.