8090 Targets the $4 Trillion Services Layer in Enterprise Software

Chamath Palihapitiya described 8090 as the first practical step toward an idea he originally called “Co-founder”: an AI system capable of helping any person create and operate a company by filling in the capabilities they lack.

The original product spec was expansive. Palihapitiya imagined someone saying, “I want to open a flower shop,” and the system spawning agents: one to analyze local demand and traffic patterns, another to contact commercial real estate brokers, another to survey the flower market, another to set up a Stripe merchant account, another to arrange point-of-sale hardware, another to manage store buildout, another to handle supply chain. For a flower shop, that might mean 10 or 20 processes. For an airplane company, he said, it might mean 10,000.

What mattered was the orchestration layer: a conductor capable of turning economic intent into coordinated execution. Palihapitiya framed this as part of a long historical trend toward more companies, from the first company to hundreds, then tens or hundreds of millions. To get from roughly that order of magnitude to “10 billion companies,” he argued, people would need something that compensates for their weaknesses and gives them the leverage of an unusually capable co-founder.

That ambition connects directly to his stated interest in economic mobility. Palihapitiya tied the co-founder idea to “giving every single human on earth the ability to be economically independent, self-sufficient, take care of themselves, take care of their family, have money to spend on the things that they want.” He called that “an extremely decentralized view of capitalism and democracy at scale.”

Jason Calacanis connected the mission to agency and self-reliance, arguing that one of America’s problems is that “self-reliance” has been displaced by “victimization” and “handouts.” For Calacanis, the practical path back is empowering more people to start their own companies. Palihapitiya agreed, but his route into the problem is narrower and more infrastructural: before everyone can have a co-founder, companies need a better machine for building software.

He said the full “Co-founder” version was not ready for the current market. Reading the spec back, he concluded that it was probably a 30- or 40-year journey. 8090, as it exists now, is the nearer-term wedge: a company built around software production for enterprises, including large and regulated organizations that need more governance than ad hoc AI coding tools can provide.

Palihapitiya said the first major premise behind 8090 came from looking at software’s economic role in the world. For purposes of the discussion, he put global GDP at roughly $130 trillion and estimated that about 90% of it now requires some form of technology or software to operate. Software, in his framing, enables around $100 trillion of economic activity.

The second premise was the cost structure of that enablement. Palihapitiya put annual software spending at around $5 trillion. He divided it into two major categories: roughly $1 trillion in licensing revenue paid to software vendors such as Workday, ServiceNow, Jira, Linear, GitHub and others; and roughly $4 trillion in maintenance, migration, and services.

That $4 trillion services layer is where he sees a large opportunity to change how enterprises build, customize, and maintain software. Once a company implements an enterprise system, it often brings in consultants to keep it running, extract value from it, migrate it, customize it, or prevent it from breaking. Calacanis described this from his own IT experience: a management decision to adopt a system such as Lotus Notes would leave implementation teams responsible for making the software actually deliver value, not the vendor.

Palihapitiya’s “lightbulb moment” came from comparing that standard enterprise software stack with the practices of companies he regards as extreme success cases: Facebook, Google, and Tesla. In his view, those companies used comparatively little of the traditional $5 trillion stack and built much of what mattered internally. He said the best companies seemed to have “an allergic reaction” to standardized enterprise software.

His analogy was a burger business. If two restaurant founders are handed the same sesame seed buns, patties, tartar sauce, and pickles, differentiation eventually requires sourcing or making better inputs. Yet in enterprise software, he argued, many companies claim differentiation while running themselves on the same underlying software as everyone else. That standardization is what allows companies such as SAP and Oracle to become enormous: they have persuaded scaled enterprises that, past a certain level of complexity, the only viable answer is their solution.

Before AI, Palihapitiya said, he would have agreed that a defense contractor or other large enterprise could not realistically build custom software the way Facebook or Tesla might. AI changed the cost curve. He saw the unit cost of software production approaching zero: no task too menial, manual, repetitive, or difficult to be addressed by models as they improved.

Calacanis brought up “vibe coding,” the practice of using AI coding systems to rapidly generate software from prompts, and noted that while some dismissed it, the tools were visibly improving every few months. Palihapitiya’s view was more skeptical: vibe coding was useful for showing “the art of the possible,” but he expected it to produce “a bunch of trash” — at best prototypes or proof-of-concept work.

The opportunity, as he described it, was not merely to let people ask AI to code. It was to create an enterprise-grade system that could bring the margin and customization benefits of internal software development to companies that historically lacked the ability, while capturing enough structured data from that process to make subsequent software safer, faster, and simpler to build.

8090’s current product is called Software Factory, a name Palihapitiya said was chosen because it honestly describes the first version of the goal: “a factory that makes software.” He compared it to a Tesla Gigafactory. Raw material enters the front; finished Teslas come out the back; between those points are logical stations where specific tasks happen.

In Software Factory, the raw material is “pure raw intention.” That intention might arrive as high-level business requirements, consulting decks, regulations, new laws, or even a consent decree from the government. Senior leaders manage risk and allocate capital toward ideas; the factory is meant to give that intent a shape and move it through a governed development process.

The first output is a detailed PRDA product requirements document that defines what a product or system must do before engineering begins.. Palihapitiya described the front of Software Factory as a “refinery”: users can dump in Zoom meetings and other materials, and the system sorts through them to produce a PRD. Humans and agents then collaborate in a shared document, debating, questioning, approving, and rejecting until the requirements are locked down.

That emphasis on documentation is deliberate. Calacanis noted that PRDs were central at AOL, and Palihapitiya used the point to argue that modern software development lost something when it absorbed the Facebook-era culture of “move fast,” “break things,” “build first,” and poor documentation. He said that culture made sense for Facebook in a specific context: it was one of many social networks, competing against established players such as Friendster and MySpace, without much money. Speed was an advantage.

His criticism is that later engineering organizations copied the slogans without the context. The result, in his view, has been weak documentation, tribal knowledge trapped in people’s heads, missing standard operating procedures, and brittle production systems.

After the PRD comes an engineering plan or blueprint. Palihapitiya compared this to handing a house sketch to an architect and asking for buildable plans that specify the foundation, wiring, and technical details. The product can incorporate system representations such as Mermaid diagrams and other artifacts needed to describe the system.

Only after that does the product extract work orders — the equivalent of GitHub issues — that can be handed to coding agents. Palihapitiya argued that AI coding agents perform well when they are directed by strict guardrails and precise specifications. Calacanis agreed, saying that the more detailed the input, the better the AI can structure it, ask clarifying questions, and identify blind spots.

8090 supports multiple coding tools through MCP, including Cognition, Codex, Claude, Cursor, and others. Palihapitiya’s view is that those vertical coding tools will converge and reach rough parity; the durable value lies in keeping the whole system synchronized.

The control plane sits above changing models and tools, allowing a company to use the best model for a task at a given moment while maintaining multi-user collaboration, governance, auditability, and control.

The forward flow is only half the system. Palihapitiya emphasized that Software Factory also works in reverse. If an engineer gets paged at 3 a.m., fixes a bug, and pushes code to production, Software Factory detects the change and propagates it backward: updating the work order, the engineering plan, and the PRD. 8090 calls this “binding”: keeping requirements, plans, work orders, tests, code, and production changes linked so that the system does not drift out of sync.

The use case he gave was a regulated healthcare company dealing with legal changes. When the law changes, the company feeds the new material into the system; Software Factory understands the difference between the current PRD and what must change to comply; humans review and approve; the change propagates into the engineering plan, work orders, and code.

For small products, Palihapitiya acknowledged, this may seem like overkill. For large enterprises in regulated markets, he argued, it is the level of governance and auditability they require. Calacanis named finance, healthcare, education, and pharma as examples where mistakes can have a large blast radius and consequences that last years. Palihapitiya added pharma explicitly.

Calacanis framed Software Factory as a third path for large companies. Enterprises often choose between buying software and hiring implementers — IBM, McKinsey, Ernst & Young, and others — to make it work. 8090, in his telling, offers a way for companies to use new AI tools to control their destiny and reduce dependence on middlemen.

Palihapitiya did not make the argument as a blanket attack on consulting firms. He said some large firms have the potential to thrive in AI, naming Ernst & Young and Deloitte as examples. His more critical assessment was about firms that choose a single model provider. In Palihapitiya’s view, it is “quizzical” when companies such as PwC or Accenture pick one of Anthropic and OpenAI, not because the models are weak, but because that decision ties the firm to one technological roadmap in a market where model providers keep leapfrogging one another.

The second issue, according to Palihapitiya, is that model companies themselves are moving toward the same enterprise ROI layer. He said the “end boss of tokens” is proving return on token spend to Fortune 2000 or Fortune 5000 CEOs — executives who need a line of sight to making more money than they spend. In his view, that is why Anthropic and OpenAI have started consulting joint ventures.

A model-agnostic control plane, he argued, gives consulting firms more optionality. Palihapitiya said 8090 works closely with Ernst & Young and Deloitte and benefits from their relationships, trust, independence, and auditability. He acknowledged his bias because they picked 8090, but his argument was that a control plane lets consultants solve client problems without betting the whole practice on a single model provider.

The sequence he laid out has three phases. For the next few years, 8090 expects to sell major transformations into large enterprises and do the “methodical work” of making them successful. In phase two, Software Factory would be “submerged under the waterline,” becoming less visible infrastructure. Phase three returns to the original “Co-founder” idea: a small interface, perhaps voice-driven, sitting on top of a robust factory that understands how to get work done and tie it to GDP and outcomes.

Calacanis compared that trajectory to infrastructure such as water: the user turns on a faucet and does not see the system beneath it. He also warned about the waste that can emerge when usage feels unlimited, comparing token usage to leaving water running or pulling a slot machine because an AI system gets 60% or 70% of the way there and tempts the user to spend again.

Chamath Palihapitiya said 8090 does not have a conventional org chart or hierarchy. He began rethinking the structure because the default motion inside the company was to ask who should report to whom, which he saw as the start of internal politics and fiefdoms. In a world of AI, he said, that did not feel like the right operating model.

His alternative is a framework he calls “System on a Chip.” The analogy is the iPhone. Inside the device are chips on a circuit board, connected by specific interconnects. The camera chip does not need to know everything the power management chip does; it needs to receive a defined signal and act on it.

Applied to an organization, each function becomes a chip defined by its inputs and outputs. Palihapitiya said he told marketing not to worry about reporting lines. Marketing has two inputs: money and content. It has one output: leads. Those leads can be enterprise leads, subscription leads for Software Factory, or job applicants, but the chip’s job is to take its inputs and produce that output. The internal design — the “cores,” “chiplets,” loops, and tactics — can then be debated.

Sales becomes another chip. Finance becomes another. In Palihapitiya’s description, the finance chip’s output is TCV, or total contract value. TCV then feeds the 8090 Enterprise team to build, and finance to account. The point is to make boundaries explicit and measurable. Agents can sit at those boundaries, observe signals, and reduce the personality-driven disputes that often occur between functions.

Jason Calacanis gave an example from his own operation: evaluating applications for early-stage startups. His team had added “coachable” as a positive scoring factor. Calacanis objected because, in his view, the difficult and uncoachable founders were often the ones who produced returns.

Palihapitiya used a similar boundary example: marketing generates a lead, sales says “the lead sucks,” and the dispute becomes emotional. In his preferred system, a signal processor at the boundary qualifies the lead as high or low quality, scores it on a spectrum, tracks conversion, and turns the discussion into a numbers problem.

He credited two sources for the organizational model. One was Jack Dorsey’s memo, which he said he initially could not understand. Dorsey’s language about world models and sales models kept turning in his head until, in a meeting with his co-founder Sina, Palihapitiya arrived at the System on a Chip idea: show the inputs and outputs, debate the pinout and interconnects.

The second source was Elon Musk’s explanation of the Tesla Gigafactory. Palihapitiya recalled seeing the layout and asking whether it was a chip. Musk corrected him: it was “the machine that makes the machine.” That phrase stuck because it moved the design problem up a level. The question is not just how to make the product, but how to design the repeatable system that makes the product.

Palihapitiya said Software Factory is built around a knowledge graph and a network effect informed by his experience at Facebook and Slack. At Facebook, he said, he helped build growth as a craft by automating, instrumenting, and applying machine learning to it. He recruited a seven-person Growth Circle, three of whom remain at the company as CXOs, including the COO and CMO. At Slack, he said, Social Capital led the Series A and he wrote a memo for Stewart Butterfield called “Intercompany Edge Effects.” At 8090, he wrote a PRD for the network effect: the ability for the n+1st piece of software to improve based on the system’s experience building the n pieces before it.

He did not disclose the full mechanism, but said it is working. In one example he cited from a third-party tweet, someone claimed to have used Software Factory to unbundle $5 billion of ISV licenses. Palihapitiya took that as evidence that the product works.

Palihapitiya said 8090 raised $20 million in its seed round roughly two years before the interview. That round included Calacanis, David Friedberg, David Sacks, Nikesh Arora, Adam D’Angelo, and several friends from the group chat, including Andrew Bogut, David Lee, Skye Dayton, and Diego Berdakin.

The new financing was discussed as the Series A. The source description identifies the round as $135 million. Palihapitiya characterized it in the discussion as “about a hundred” million dollars after the earlier $20 million seed. He said the A was led by Marc Benioff and Salesforce Ventures, and also named Thomas Laffont at Coatue, Yuri Milner, and Xander Lurie among the new investors.

Calacanis asked what Palihapitiya is learning as a CEO versus a capital allocator. Palihapitiya’s answer was that the job is fundamentally the same, but the scope of allocation is broader. As an investor, he said, he was allocating one of five units: money. As a founder CEO, he is allocating all five: time, reputation, social capital or influence, human capital, and capital.

The work carries a different emotional load. Palihapitiya said his current state is “this constant state of worry”: he does not want to let friends, investors, or employees down. He connected his capacity to absorb that anxiety to a difficult childhood, including an alcoholic father and physical violence. He said he holds no grudge, but that the environment trained him to absorb chaos.

His time now goes into founder-led sales, product management for 8090’s network effects work, and organizational design. Calacanis argued that founder-led sales keeps the CEO close to truth: sales teams can distort reality for both the founder and the customer, while a founder in the room hears the market directly. Palihapitiya said plainly: “I sell the enterprise software now. That’s what I do.”

Palihapitiya used the same five-part capital framework to explain why he started businesses around research and wine. Learn With Me, his research community, grew out of his desire to keep a “prepared mind” for allocating capital. He said he once hired a research service described to him as the team that teaches Bill Gates when Gates wants to learn something. He asked them to teach him about energy because he wanted to understand climate change, energy production, AI silicon, and power. The work was valuable, he said, but cost him “many millions” for a few months. He then created an internal research team for himself.

The quality-control problem was how to know if the content was wrong or weak. His answer was to make it a paid subscription community. Learn With Me now has many thousands of subscribers paying about $1,000 a year, and churn is the simplest value signal. If people churn, the content is bad; if the service grows, it suggests the work is timely and accurate.

Calacanis called this a “Tom Sawyer version of entrepreneurship”: turning a cost center into a community and business whose users validate and improve the output. Palihapitiya said the team profits from it, though it is not where he defines his upside.

Drink With Me followed a different logic but a similar pattern: productizing something already meaningful in his life. Palihapitiya framed wine as a quality-of-life choice and criticized a middleman economy that marks up bottles, creates gatekeeping, and can separate artisans from the people who value their work. His answer was to get a liquor license, let a community buy with discounts he can access, and build relationships with winemakers who may eventually face succession questions.

Learn With Me, Drink With Me, All-In, and 8090 are all, in Palihapitiya’s account, different expressions of a late-40s shift. After Facebook, he had allocated money, reputation, influence, and human capital, but had not put all of those together into one company. 8090 is that allocation of all five forms at once.

When Jason Calacanis asked how Chamath Palihapitiya is preparing his children for AI, Palihapitiya said he did not have a good answer. He said he asks the question more than he answers it. The only conclusion he has reached is that it is important for his children to have “their own adventure”: to do something interesting, have wins, have losses, and experience risk and agency directly.

That answer rhymes with the agency premise behind the co-founder idea, though Palihapitiya did not present it as the same argument. In both cases, the concern is what people can do when more of the environment around them is mediated by AI or made more abundant. For enterprises, his answer is a governed system for creating software. For children, his answer is less precise: they need an adventure of their own.

Palihapitiya argued that even if food, shelter, and basic necessities become more abundant, human physiology will not change in one generation. The human desire for agency, risk, and learning is hardwired. Calacanis added problem solving and socialization: people want to play games, solve problems, and interact with others. If those needs are not met, he argued, society will medicate the resulting dysfunction rather than address the absence of meaningful challenge.

Palihapitiya said “adventure” was the best word he had found because it captures the whole set of needs. He sees his own life as an adventure, still “only half done,” and said the same of Calacanis’s life. Calacanis offered a related term: exposure. His goal is to expose his children to possibilities and high-agency people early enough that something can pull them forward. He said he wished he had been exposed earlier to entrepreneurship and risk-taking.

For Calacanis, that means taking his children into environments where they can see ambition up close, including Founder University in Tokyo. He contrasted that with his own childhood, when air travel was financially out of reach and family vacations were constrained by how far the van could reasonably drive before breaking down. His point was not that children need a specific destination; it was that early exposure to possibility can change the set of risks a young person is willing to consider.

The discussion of children connected to Palihapitiya’s definition of success. Calacanis brought up Thomas Keller and Warren Buffett as examples of people who could have stopped long ago but did not. Palihapitiya said he deeply admires the “core group of maniacs that never stop.” His explanation was that they are playing an internal game. If life is defined by external metrics, eventually a person hits the number or status and stops.

Calacanis closed the thread with a practical parenting example: helping his daughter try skills that take several attempts before the “unlock,” such as skiing or pickleball. The point was not the sport itself, but the repeated experience of being bad at something, returning to it, and discovering competence after the third or fourth try. For him, that builds resilience and appetite for the next hard thing.