OPC is the new startup

Last week, in California, and this week, here in China, I heard a lot about OPC, or One Person Company. During the visit to Alibaba, the topic came up as part of a discussion about artificial intelligence agents and the future of business. The question goes beyond entrepreneurship alone: how much can one person build and operate before needing to set up a larger organization?

Researching a market, developing an application, preparing a campaign, analyzing data and organizing processes. Activities that used to require different professionals are starting to fit into a much smaller operation. This doesn't mean anyone can do everything well. It means that some limitations that once seemed natural are no longer so definitive.

This leads me to a provocation: OPC is the new startup. Not because startups are going away, but because we are changing how much one person can build before needing to construct a larger structure.

Each generation has lowered a barrier

Running a business alone is nothing new. Independent professionals and solo businesses have existed for a long time. What changes is the set of resources available to those starting out.

The internet expanded access to the market. The cloud made it possible to rent infrastructure as needed, without building an entire physical structure before testing an idea. The launch of Amazon S3, in 2006, is one of the milestones of this infrastructure-as-a-service logic. The original AWS announcement already presented storage billed by usage.

Then, APIs, subscription software and automation tools made it possible to combine ready-made services. A company no longer needed to develop every component of its own operation.

Startup culture also helped consolidate another order of execution: test a hypothesis, put a first version into use, listen to the customer and adjust before investing in a larger structure. Lean digital businesses and MicroSaaS expanded the possibility of addressing a specific problem, generating revenue and keeping a small operation by choice.

These models coexist. There is no mandatory sequence in which each new approach replaces the previous one. But there is a common direction: reducing the resources needed to get started. Artificial intelligence adds another layer because, besides offering tools, it can take part in actually doing the work.

From operating tools to coordinating work

For a long time, gaining productivity meant using better software. Someone was still needed to feed the system, interpret information and execute the next step. With agents, part of these steps can be connected. The technology can search for information, use tools, produce a deliverable and send it for review.

When different agents split up functions, the discussion stops being just about speeding up a task. It becomes about organizing a workflow, with planning, execution and review. It's this shift that helps explain OPC: one person coordinating a delivery capacity larger than they could execute directly.

During the visit to Alibaba, an example was described that helps visualize this shift: connecting requests received by email to supplier research, proposal preparation and sending responses to customers. What's interesting is looking at the whole sequence, not just the speed of writing a message.

This business account, on its own, doesn't prove how much time was saved or whether quality was maintained. But it makes the discussion more concrete: activities that are normally distributed among different people and systems can be coordinated in a single workflow. Someone still needs to define the selection criteria, check the proposal and approve what reaches the customer.

The vision of agents interacting with other businesses' agents, called agent-to-agent or A2A, expands this possibility. There are still challenges around integration, trust, authorization and accountability. We are not looking at a fully resolved commercial autonomy. The very conversation about the solution presented brought up limits on direct payments and compliance requirements still under development.

The data show a change, but require caution

According to Carta, the share of companies with a single founder in its base went from 23.7% in 2019 to 36.3% in the first half of 2025. The survey uses data from tens of thousands of companies in the United States, it doesn't represent entrepreneurship worldwide. The numbers are in the Solo Founders Report 2025.

It's a significant change. But solo founder doesn't mean a company without employees. The same report shows that companies with a single founder hire their first employee sooner than companies with multiple founders: the median was 399 days, versus 480 days.

So, these numbers can't be used as proof that we're heading toward companies without teams. Nor can we attribute all of this shift to AI. What they show is that starting with a single founder has become more common. The decision to expand the team still exists.

There is also evidence of increased individual capacity. An experiment with 791 professionals from Procter & Gamble, published in 2026 in Organization Science, compared people working individually or in pairs, with and without AI, on product innovation challenges. Individuals with AI achieved performance similar to pairs without AI. The published version of The Cybernetic Teammate details the results.

The study doesn't demonstrate that one person can run any company alone. It evaluated specific tasks, not the complete operation of a business. Even so, it helps explain why activities that once required complementary skills can now be carried out differently. The gain is not just in doing things faster, but in accessing knowledge that improves the quality of delivery.

In China, the individual company is also a collective discussion

An interesting part of this conversation here in China is that OPC already appears in initiatives supporting entrepreneurship. In April 2026, Xinhua described local programs involving communities, computing capacity and support services. The report cited, in Guangdong, goals of creating ten reference communities that year and a hundred by 2028. These are announced goals, not results already achieved. The overview of OPCs published by Xinhua helps put this movement in context.

A one-person company is not a person working isolated from the world. It depends on AI models, infrastructure, suppliers, payment methods, distribution channels and relationships with other people. Part of the capacity that used to need to be inside the company is now accessed outside it.

This is a reading that interests me especially because I work on ecosystem building at NewHack. The more capacity an individual gains, the more important the environment that allows turning that capacity into a business can become. Technology reduces some dependencies. It doesn't eliminate the need for connection.

Another example from the visit was 1688, presented by the team as a B2B platform that connects buyers to suppliers. Cases were cited of factories with products in stock and small orders, without this being a universal condition for every supplier. It's a complement to the OPC discussion: the capacity to build also depends on access to the supply chain, not just on software.

Building got more accessible. Finding a market is still hard.

This is where I would be careful about the enthusiasm. A working application is not necessarily a business. A generated campaign is not distribution. An agent responding doesn't mean the customer was well served.

When building becomes more accessible, differentiation needs to show up elsewhere. Deeply knowing an industry, having access to customers, earning trust and understanding an operation make even more of a difference. The advantage may lie less in being able to develop a feature and more in knowing which feature deserves to exist.

This opens space for people who know a problem very well but previously didn't have the resources to turn that knowledge into a product. At the same time, it increases competition. If I can quickly launch a solution using tools available to everyone, other people can too.

And the operation needs to add up. AI has a cost to use, can make mistakes and requires supervision. An interesting demo can turn into an expensive or unreliable service once real customers arrive. The math needs to account for the delivery the customer accepts, including review and rework.

There is also the risk of building a company where everything still depends on the founder. If every new customer requires a few more hours from the same person, maybe the limitation has just moved. Having several agents doesn't solve this automatically. You need to watch what they deliver and what happens when some step fails.

OPC can be a starting point, not an obligation to stay solo

Startup and OPC describe different things. In Paul Graham's definition, a startup is a company designed to grow fast. OPC, in this discussion, is about the organization of work. A company can have both characteristics. This distinction comes from Startup = Growth, published in 2012.

It can start with one person and then hire. It can stay small and profitable. It can seek investment or grow on its own revenue. I don't see this as a competition to find out who can hire the least. I see it as an expansion of choices.

Hiring because there's a clear need. Seeking capital to accelerate something that has proven its value. Building a structure that keeps pace with the business, instead of needing to build that whole structure to find out if the business makes sense.

For entrepreneurs in Brazil, this possibility deserves attention. People with market knowledge, operating experience and access to real problems can find new ways to build. OPC is the new startup is a provocation about this shift, not a prediction that every company will have the same shape.

For a long time, the question for anyone who wanted to start a business was: what team and how much money do I need to get started? Now, in a growing number of businesses, we can ask a different question: how much can I learn, build and validate before needing that structure?

The company can start smaller. The ambition doesn't have to.