No sales, doubled revenue, rejected financing: a 14-person team's "anti-consensus" experiment in the AI wave

From execution to judgment, and then to imagination.

In the first half of 2026, NocoBase made a decision: to stop expanding its headcount.

The company now has 14 employees. Over the past few years, the team has roughly maintained a pace of adding one or two people each year. As the product needs to cover more and more features, the R&D team is often overloaded, and some demands have to be queued. If this pace continued, the team would have gradually grown in size.

AI interrupted this rhythm.

Zhou Yanliang, the founder of NocoBase, and his team made a rough estimate that the current product R&D efficiency is about 3 times that of the past. The vast majority of code is generated by AI, and humans barely write code manually anymore. The backlog of work that used to pile up in the R&D team has started to be cleared, and the necessity of simply recruiting more programmers has declined accordingly.

At a team gathering in the first half of this year, they decided to keep the team size at its current level. If someone leaves, a replacement can be hired; if an existing position is no longer suitable, adjustments can be made. But there is no new task that requires hiring more people to solve.

"If we expand the headcount to 10 times the current size, I have no idea what those people would do," said Zhou Yanliang.

NocoBase's product was initially developed by a team of three. It started as an open-source no-code development platform for professional IT teams. Its commercial version was officially released in 2024, and its revenue exceeded 10 million yuan in 2025. It is understood that as of the interview, NocoBase's revenue in 2026 has at least doubled, with domestic and overseas revenue each accounting for half.

AI's real integration into the core R&D process happened later.

After the second half of 2025, the capabilities of AI coding tools such as Claude Code improved rapidly, and NocoBase experienced a clear internal upheaval. Some even judged that there might no longer be a need to continue developing no-code platforms — since AI can write software directly, is such an intermediate layer still necessary?

A few months later, the product development continued, and the team did not shrink.

Changes took place inside the organization.

In the past, the scarcest resource was R&D time. Now, some R&D personnel have spent about half of their time working on customer projects. Technical support staff can directly participate in fixing product issues, and the once clear boundary between development and implementation has begun to blur.

As execution becomes increasingly easier, the company's attention has shifted upstream. They now spend more time discussing one thing: what exactly should be built.

Make the company lean first

NocoBase has an office in Beijing, but few people go there.

Team members are distributed in different cities, there is no clock-in requirement, and no fixed office location. There is a 20-minute morning meeting every day, and a slightly longer weekly meeting every Monday. Apart from that, most communication is done asynchronously.

They use Feishu, but they believe that the most important thing is not chatting, but documents.

When a problem arises, the person concerned usually thinks it through first, writes down the problem, judgment and possible solutions clearly, and then submits it to others for discussion. General information is directly synchronized in text, and only issues that are really worth discussing are brought up in meetings.

The team meets twice a year. Once in May, once in November, they stay in one place for 7 to 10 days each time. For the rest of the time, everyone lives and works on their own.

This working method did not emerge after the AI wave broke out.

Zhou Yanliang participated in the founding of remote work communities very early. When NocoBase was established, it naturally adopted this model.

In his view, remote work is not suitable for many people. If a person needs a clear office scene, attendance check and task assignment to restrain himself, he will easily lose his rhythm in such a team.

Therefore, NocoBase spends a lot of time on recruitment.

In their view, recruiting each team member is like "falling in love". Some candidates will communicate for half a year before finally joining. Before the formal recruitment discussion, the candidate has usually studied NocoBase on his own, formed some judgments about the product, and even written articles or complete analysis reports. They value these qualities more than what specific technologies the candidate masters.

From NocoBase's perspective, what they are looking for is not just people with strong pure capabilities. More importantly, whether their way of working and thinking habits fit the team. Some people have excellent professional capabilities, but if they are used to relying on clear division of labor and external constraints, they may not be suitable for such a small team with high autonomy and long-term asynchronous collaboration.

NocoBase now has 14 people: one for marketing, one in charge of partners, one for solution design, one for technical support, one for finance, and the rest are mainly in product and R&D roles. There is no sales team.

The organization has always been very lean.

This "leanness" in the past was more of an active choice. The company does not build a large sales team, does not maintain a large-scale implementation team, does not pursue rapid headcount expansion, and has no quarterly growth requirements brought by financing.

After AI was introduced, this structure became even easier to scale.

A company that originally had very few coordination levels will see changes transmitted directly to the work itself without going through complicated organizational transformation, when the execution capability of each individual suddenly increases several times.

In the past, NocoBase used to add one or two people every year.

This year, for the first time, they feel there is no need to do that anymore.

Coding gets easier, "thinking clearly" becomes harder

AI once prompted NocoBase's internal team to seriously discuss a question: does this company still need to exist?

In the second half of 2025, AI Coding capabilities improved rapidly.

A colleague in the team once said that after using Claude Code recently, he felt that products like NocoBase "might not need to be developed anymore".

This judgment is not hard to understand.

What NocoBase used to do is to lower the threshold of enterprise application development through no-code and low-code capabilities.

Now users can describe a requirement to AI, and directly generate interfaces, data models and even complete applications. If the capabilities continue to improve, no-code platforms will easily become a redundant intermediate layer that can be bypassed.

The internal team quickly split into two opinions.

Some people think that changes will come very soon, even the number of programmers may decrease in the future, and the existing product roadmap needs to be completely adjusted.

Others think that AI is far from mature enough, most enterprise users cannot make good use of it, and there is no need to frequently change direction following technology hotspots.

This contradictory state lasted for several weeks.

Later, everyone discussed NocoBase's reason for existence seriously for several times: why are enterprises willing to pay for a software product? After AI can generate code, what problems are still unsolved?

Everyone's judgment did not stop at the model capabilities.

When an enterprise buys a piece of software, it rarely does so just to "own a piece of code". Issues including permission design, data organization, process arrangement, long-term system iteration, connection with existing systems, and maintenance after malfunctions will not disappear automatically just because a CRM Demo is generated in a few minutes.

NocoBase then began to adjust its direction.

While maintaining the current no-code version, the new version to be released in October has fully shifted to providing infrastructure for AI development, offering AI-driven delivery processes from requirements to coding, testing and release, as well as basic capabilities including data sources, data models, permissions, workflows, AI capability access, knowledge base and pluggable runtime, so that Coding Agent only needs to focus on business requirements, and help teams deliver reliable business systems faster.

The product roadmap is changing, and the R&D method is becoming more efficient.

Now the vast majority of the team's code is generated by AI. "There is barely any need for humans to modify it manually".

The remaining work is mainly in the upstream stage.

Work including architecture design, product positioning, and deciding whether a certain feature should be added to the mainline is still completed by humans.

There is a frequently mentioned saying inside the team: "Disconnect from the Internet when thinking, and stay connected when working".

AI can help look up information, discuss specific technical issues, generate code, and process a large number of documents; but the initial judgment of product positioning cannot start with asking AI.

"What value the product provides to users today", in NocoBase's view, requires the team to go to the customer's site, observe how the real business runs, and then make judgments on their own. AI can generate dozens of answers in a row, but if this kind of problem is completely handed over to the model, the product value will be greatly reduced, or even worthless.

This is very different from software development a few years ago.

In the past, whether a direction was thought through could sometimes be covered by implementation costs. It took dozens of people several months to develop the product, and then they could know whether users would pay for it after the product was finally launched.

Today, the implementation cycle is getting shorter and shorter, and wrong ideas can also be quickly turned into a complete product.

AI reduces the cost of trial and error, and also makes judgment needed more frequently.

The bottleneck NocoBase is currently facing has therefore gradually shifted away from the R&D team.

"What is more difficult today is to think clearly about the product, think clearly about the direction, and design the architecture well. The delivery speed is never a problem."

Programmers go to the customer site

After the R&D team was no longer so overloaded, NocoBase did not let these people develop more features.

Several core members began to go to customer sites.

In the past, NocoBase basically did not provide customized development or delivery services. After customers purchased the License, they mainly relied on their own IT teams to complete the deployment. A customer was usually not assigned a dedicated pre-sales, implementation or customer success team, and problems were solved through the ticket system.

This was an important reason why the company could maintain a team of more than ten people in the past.

The situation began to change this year.

Some R&D personnel spend nearly half of their time participating in high-quality customer projects. They go to the site, build systems with NocoBase, solve actual business problems, and then bring the problems found on site back to product development.

This is very similar to the much-discussed FDE nowadays.

However, NocoBase did not do this because it planned to transform into a project delivery company, as the relative value of implementation has also decreased.

According to NocoBase's estimation, a traditional code project used to take 6 months to complete, but using no-code tools can shorten the cycle to one or two months. For suitable projects, with the help of AI now, a usable version can be generated in just two days.

In the past, both development and implementation required a lot of manpower, so it was unrealistic to assign the two tasks to the same person. Today, the execution costs of both sides have decreased at the same time.

An R&D staff member participating in a customer project can directly build the system after understanding the requirements; if he finds that the product lacks certain capabilities, he can go back and let AI modify the code. The work that used to be split into several positions has now returned to one person.

Similar changes are also taking place in the technical support position.

The technical support colleague at NocoBase did not originally have a technical background. In the past, when a user reported a Bug, he needed to assign the problem to the R&D team, wait for the scheduling and fix, and then follow up the version update.

Now, he can fix some problems by himself directly.

NocoBase does not specifically require everyone to learn a certain AI tool, nor does it set indicators such as Token usage. Changes are taking place little by little in specific work.

The originally clear division of labor among different positions has begun to overlap.

Roles including product, development, implementation and technical support still exist, but one person can take on more and more work. Tasks that used to be transferred between different positions can now be completed directly by the same person.

It is still hard to draw a conclusion where this change will lead the organization of software companies to.

In the past few decades, the expansion of software companies' scale was usually accompanied by increasingly fine professional division of labor, for example, product managers define requirements, programmers develop, testers are responsible for quality, implementation staff are responsible for launch, pre-sales staff are responsible for solutions, and customer success staff maintain customer relationships.

This structure is largely built on a premise that professional execution capabilities are expensive.

When a person has more and more execution capabilities that can be called at any time, some work that had to be split in the past can be recombined.

NocoBase is now a very small sample.

Developers getting closer to users is not because the company cannot afford an implementation team. They found that coding no longer takes up most of their time, and what really needs to be supplemented is the understanding of the actual business scenario.

NocoBase believes that the ceiling of a small team now more likely lies in "imagination". No matter how many people you have, if no one knows where the product should go next, it is hard to solve the problem.

Not expanding is also a way of growth

NocoBase has never had a sales team, and customers usually find the product on their own first.

Because NocoBase is open source, and the official website also provides a trial environment, some customers have used the product for several months on their own before contacting the company for the first time. When they decide to make a purchase, they will ask how to buy, how to pay, and how to sign the contract.

The company does not have sales staff to follow up these leads continuously.

Some customers will disappear after their first inquiry, and come back several months later. Viewed on a monthly basis, the revenue fluctuates greatly; when extended to the annual dimension, the growth becomes obvious.

They are not going to change this model.

NocoBase hopes that customers will look at the product, try the product, and then judge whether it is suitable for them. The same is true for target users: NocoBase mainly serves enterprises with professional IT teams. If an enterprise does not have IT capabilities and only wants business personnel to build applications directly, they will recommend customers to choose other products.

This model naturally narrows the market scope and also reduces its own organizational complexity.

Customers usually learn about and try the product on their own first, most projects do not require a dedicated implementation team, and after-sales services are mainly completed through a unified technical support process.

This standardized model allows NocoBase to serve widely distributed customers with a very small team. At present, its customers cover at least 60 countries and regions, with scattered industries, enterprise scales and usage scenarios, and no obvious concentration in a certain type of market.

NocoBase also does not split sales revenue to each position as KPI.

The revenue target is mainly undertaken by the founder himself.

The company's revenue, customers, orders, payments and costs are all open to the team. Because of the profit distribution plan, everyone can see how much money the company makes in a year, and know how these figures are related to their own income in the end.

NocoBase rarely manages employees through attendance checks and processes, but has clear requirements for results.

It's okay to be in a bad state for a short time, but at the key node, your own work must be completed on time. The team only has more than ten people, everyone's performance is very clear, and people who cannot keep up with the rhythm for a long time cannot be hidden by the organizational hierarchy.

NocoBase also actively reduces external growth pressure.

The company has no financing, the founder provided cash flow in the early stage, and it had revenue soon after commercialization. Without the growth requirements from investors, the team does not need to increase headcount, customer scale and revenue targets synchronously for the sake of