How Blank is your Sheet?
There will be AI-native companies, and there will be everyone else
The Gap That Won't Close
Earlier this month, I attended the Acadian Ventures Summit in Carlsbad. A great time! Among many other insights that I can credit to the brilliant minds in the room, the conversations gave me a clearer picture of what is actually blocking AI-native operation inside established companies — and it reminded me of a pattern we have seen before.
When cloud computing arrived, the companies without legacy infrastructure had a real advantage. Cloud-native startups moved faster, scaled cheaper, and operated in ways that traditional companies could not easily replicate. But the laggards had time. Digital transformation became an industry. A whole ecosystem of consultants, integrators, and vendors emerged to help established companies make the journey. It took years — sometimes a decade — but most got there. The gap was real but crossable, and the people who specialized in crossing it built valuable careers doing so.
The comparison with AI is valid up to a point. AI-native organizations have the same structural advantage that cloud-native companies had. Thomas Otter, one of the sharper observers of the enterprise software world, described the blockers at the summit precisely: legacy systems that were never designed for agent access, APIs that predate the modern internet, data that is often 50% accurate at best, incumbent vendors who will control and toll the channel through which agent access must flow. On top of the technology constraints sit compliance requirements nobody has adapted for autonomous agents, management layers with rational reasons to resist change, and organizational immune systems that have had decades to develop.
But this time the gap is different in kind, not just degree.
The productivity gains available to a truly AI-native organization — the kind Pete Koomen recently described YC building over the last eighteen months — are not incremental improvements on what established companies can achieve. They are a different way of working altogether. An organizational brain that compounds. Collective intelligence that improves itself. Questions that previously required days of coordination answered in seconds. It will not feel like being slightly ahead on a digital transformation journey. It will feel like a different era entirely.
And unlike cloud, this gap may never fully close. The structural dependencies — on legacy systems, on vendor API roadmaps, on compliance frameworks written for human decision-makers — are not problems that a determined transformation program will eventually solve. Some of them are permanent features of what those organizations are.
For founders building something today, this is the argument for taking the blank sheet seriously. Not as a nice-to-have, but as a once-in-a-generation opportunity to build in a way that established organizations will not be able to replicate.
Garry Tan put it plainly on the Light Cone podcast: founders willing to build truly AI-native organizations right now are living in 2028. The cost of tokens will keep falling. What requires significant investment today will be commoditized in a few years. The window to leapfrog is now, before the approach is so well understood that doing it well is no longer a source of advantage.
For people choosing where to build their careers, the calculus is worth thinking about carefully. There has always been interesting and rewarding work in leading transformation at legacy companies. Those people develop expertise that many organizations need, and they tend to do fine. That remains true.
But the career opportunity cost of not going to work for an AI-native organization is higher this time than it was during cloud or mobile or any prior wave of digital transformation.
The gap between what it feels like to work inside an organization that has genuinely made the leap and one that is still trying to cross the distance is going to be wide. Wider than anything we have seen before in a technology transition.
It is worth knowing that before you choose which side of it you want to be on.
What YC Built
YC has spent the last eighteen months building this way internally. Not as a pilot. As the actual way they run the organization. Pete Koomen, the former founder of Optimizely who now leads their agent infrastructure, described it on the Light Cone podcast in enough detail that the picture becomes concrete rather than theoretical.
Three things they built matter most.
The first is a shared context layer. Everything that matters to YC — every company funded, every founder, every financial transaction, every partner note — lives in a single database that agents can query freely. Any question about the organization can be answered in seconds. And because the cost of asking dropped to almost nothing, the number and complexity of questions exploded. Questions nobody bothered to ask before — because getting an answer meant finding the right person, waiting for their bandwidth, translating the request — now just get asked. Constantly.
The second is self-improving skills. A skill at YC is a prompt that encodes how the best people in the organization do a specific thing. One partner wrote a skill for producing the two-sentence company descriptions that every YC partner writes constantly. Then another partner fed meeting transcripts of founders practicing their pitches into it. The skill read those transcripts and improved itself. It now performs better than any individual partner does. The collective intelligence of the organization is embedded in something that compounds without anyone tending to it.
The third is a tool registry — more than 350 tools built by people across the organization wherever they encountered work that could be improved. Not by engineering request. By the people doing the work. Koomen described this as what turns generic agents into something useful at work. The infrastructure for organizational intelligence grows wherever people are, not only where engineering capacity allows.
The result is something that looks less like a company that adopted AI and more like an organization with a different relationship to knowledge and capability altogether.
Garry Tan’s word for it is superintelligence — not in the science fiction sense but in the practical sense that the organization can do and know and improve things that no collection of individuals working conventionally could match.
Why Most Companies Won’t Get There
The honest version of why is not that their people are incapable or their leaders aren’t serious.
It is that the organization was built for a different world, and the assumptions baked into its structure have stopped feeling like assumptions. The finance department is organized this way. The sales team is organized that way. The management layer moves information between the people doing work and the people deciding about it. None of these were arbitrary choices — they were reasonable responses to real constraints: expensive communication, limited information storage, the impossibility of one person knowing what many people knew. The constraints dissolved. The structure stayed.
Adding AI to that structure produces real gains. But there is a difference between a company that has added AI to how it works and one designed from the beginning around what AI makes possible. That difference is not incremental.
There is also a cultural constraint that Garry Tan named directly. At YC, every agent conversation is visible to every full-time employee. That decision required believing that transparency produces better outcomes than control. Most organizations don’t make that bet — they lock down context, restrict access, manage information flow, for reasons that are often sensible. But those restrictions also mean the shared context layer never gets built. The organizational brain stays fragmented in individual heads. AI makes individuals slightly faster without making the organization fundamentally smarter.
Remote work had the same dependency. The organizations that adapted to it most fully were the ones already extending trust to individuals — evaluating results rather than presence.
The ones that treated it as a surveillance problem struggled in ways that had nothing to do with the tools. AI transformation will play out the same way.
The tools are available to everyone. The organizational culture that lets them work at their full potential is not.
What Organizations Actually Are
Stripped down, an organization is a collective attempt to produce outcomes no individual could produce alone. The hierarchy, the departments, the job titles, the meetings, the processes — none of these are the organization. They are scaffolding that accumulated as people tried to coordinate, share information, and make decisions at scale. Useful scaffolding. Built for real constraints. But scaffolding nonetheless.
When those constraints dissolve, the scaffolding becomes optional. And optional scaffolding treated as mandatory is dead weight.
Garry Tan offered a recharacterization of what Jack Dorsey is trying to do at Block that captures this precisely. Not a company that has reduced its workforce or added AI features to existing processes — a focused intelligence organized entirely around a specific outcome. Enabling people anywhere in the world to move money. Not a company with a payments division. A payments intelligence. The structure is whatever serves that purpose, and nothing else.
Most organizations have never made that shift because the how and the what got tangled together decades ago and calcified. The how became infrastructure, became headcount, became institutional knowledge embedded in roles and relationships that couldn’t be extracted or transferred. When institutional knowledge can be captured, shared, and built upon systematically — when capability doesn’t have to live exclusively in individuals — the question becomes available in a way it wasn’t before: what are we actually trying to produce? And then: what would an organization designed purely to produce that look like, starting today?
For founders building something now, that question is urgent in a way it has not been before. The blank sheet is actually there.
The Choice
The default, even for startups, is to keep doing what we’ve always done. Hire a head of finance and build out a team. Structure a sales org that looks like the sales orgs that have worked before. Put management layers in place because that is what organizations do.
The necessity behind those choices is gone. The habit remains.
What YC demonstrated — what Dorsey is attempting at Block in a more radical form — is that it is possible to start from the outcome and work backward to the structure.
The questions worth asking at the founding of an organization right now are not the conventional ones.
Not what functions do we need, what roles do we fill, what does the org chart look like at fifty people. But: what are we actually trying to produce? What does an intelligence organized purely around producing that look like? What is the minimum structure required to hold it together and let it compound?
Existing organizations don’t get to ask these questions seriously. The weight of what already exists is too heavy.
Founders get to ask them. The ones who build their answers into the foundation — rather than inheriting the scaffolding and hoping it produces the outcome — are the ones for whom the window Garry Tan described can be completely transformative.
The way a company works has always been a potential source of advantage. Fully distributed and officeless still is — I believe that as strongly as ever, and the companies that have genuinely figured it out have something most of their competitors cannot easily replicate. But fully distributed, as hard as it is to do well, is still a different way of organizing people to do familiar work. Okay, maybe with some genuine accelerations like asynchronous.
But AI-native is something categorically different — not a different way of organizing people, but a different relationship between people, knowledge, and capability altogether.
Founders have always told themselves they had a blank sheet. For most of the history of company building, that was only partly true — the habits, the templates, the inherited assumptions about what an organization looks like all filled it in before the first hire was made. Now, for the first time, the sheet can actually be blank.
How blank is yours?
Sources:
Pete Koomen and Garry Tan, The Light Cone podcast, Y Combinator, May 2026. Thomas Otter, “APIs, Hormuz, Headlessness and the Incumbents,” Work in Progress, May 30, 2026. thomasotter.substack.com. The characterization of Jack Dorsey’s work at Block draws on Garry Tan’s framing in the Light Cone conversation.