Six Years of Revenue, Then Four Times That in Eight Months

Sequoia's Training Data hosts George Robson and Pat Grady introduce Jan Oberhauser as one of the sharper growth stories in automation: n8n added, in eight months, about four times the revenue it had taken the first five or six years to earn. Sequoia has been in the company since 2020, not long after Oberhauser, then a solo founder in Germany, shipped the first version in 2019. The product began as a developer-leaning workflow tool — nodes wired to nodes, self-hostable, in the same neighborhood as Zapier — and the hosts' framing is that the explosion came from turning that canvas into an orchestration layer for AI-powered applications.

Robson opens with the obvious question: what changed? Oberhauser says the answer is not the last eight months. It is two decisions the company made almost two years earlier. One was a bet on AI. The other was a stranger marketing call: stop optimizing for sales leads. Before he unpacks either, the episode's cold open already states the product thesis. There is no default n8n stack. People plug in any LLM, any memory or vector store, any application. He does not know whether one model wins, a million small ones do, or something in between, and n8n does not have to care.

Later he will put the same idea in a sentence the hosts had been waiting for: n8n is, by now, probably the easiest and most powerful way to build AI agents. You go from a first idea to a prototype to production without writing the Python that used to be required. The connect-everything-to-anything line is not a slogan for him. It is the reason the company can sit still while the model race thrashes overhead.

When the wave started, the team was not calm. Oberhauser says they knew AI would mean one of two things for n8n: a huge opportunity, or the demise of the company. They looked at what others were doing and saw a pattern — nice AI features bolted onto existing apps. That, he decided, was not the kind of move that keeps a company around. What mattered was becoming part of the value chain. So n8n did not just add AI. It let people build AI-powered applications on the product itself.

The second decision lived in marketing. The team had always carried a dual goal: generate leads, and drive adoption. Adoption was consistently strong; leads fell short. So marketing pulled people off adoption work to hit the lead number, and long-term plans evaporated. Oberhauser's fix was blunt. The company already had a community and bottom-up usage. Instead of forcing a motion that was not true to n8n, they removed the lead goal and swapped it for adoption inside large organizations, then went all-in on community: more events, more content, more people actually using the product.

He warns that this kind of change shows nothing for months. You have to trust it and keep doubling down while it matures. The flywheel showed up around the end of the previous year, in December: the market finally registered that n8n had become an AI tool, and YouTube filled with tutorials. More content ranked better, which pulled in more creators, and things started to explode. Robson had heard a board-meeting story that Oberhauser coded the first AI-nodes product on a long flight home from San Francisco. Oberhauser laughs it off. He has built plenty of things on planes and trains, but not that one, and he no longer ships production code in the air — only prototypes. Dodgy Wi-Fi, Grady suggests. Maybe, Oberhauser says.

Robson takes him back to 2019, a pre-seed founder writing the product alone. What has stayed true, Oberhauser says, is community. In the first week that April, one of the earliest hires was in developer relations, because he already believed n8n would not become the company he wanted unless he invested in community immediately. The other constant is candor, and the license is the exhibit. n8n has never called itself open source. The source is public, anyone can use it for free, including in production, at home or inside a large organization. What they cannot do is take the code and sell a hosted n8n, or stand up a competing commercial automator on top of it.

He chose that design after watching open-source companies change licenses midstream and watch their communities explode in anger. The interesting part, to him, was that people were not mainly furious about the new license. They were furious that the company had changed the rules. So he said the quiet part on day one: he is not giving n8n away because he is a good person. He wants to build a business around it. He wants to get paid, and he wants everyone else on the team to get paid, because he thinks that is in everybody's interest. Six years on, that speech has not changed.

Tactically, the rule is to take people with you. Early n8n collected no telemetry at all, which was fine at a tiny scale and fatal once the company needed data to improve the product fast — the difference, he says, between a company that survives and one that does not. They added telemetry. They also added paid features. Every time they made a bigger change, they posted to the community forum first, explained what they wanted to do and why, and listened. Hide the rule change and you get the license revolt. Narrate it and you keep the room.

The person who made the ethos concrete is Ricardo, an engineer in Florida who found n8n on Product Hunt and started writing integration nodes. One node, a thank-you and some feedback. Then two, then three. He kept going until he had built fifty or sixty. As soon as Oberhauser had the first real money in the bank, he hired him. Show people you care, take them with you, and they return far more than you gave. He will not brand n8n as an open-source company, but he runs it with that give-first ethos, and he points at the unpaid tutorials now flooding YouTube and LinkedIn as the return.

Grady asks him to rewind to the make-or-break AI moment. One image stuck: a Pinecone funding announcement, a round Oberhauser remembers as being on the order of a hundred million dollars, not long after another raise. He wondered why the market had suddenly cared so much, then realized Pinecone had stopped being "a vector database" and become a deep database for AI. The identity had moved. n8n, he thought, had to make a similar move.

Until then, n8n had done what most competitors still did: an OpenAI node that fired HTTP requests. You could build some nice use cases with it, and you still can. It was not enough for the powerful ones. You needed real agents — changeable prompts, tools, a vector database, output parsers, the whole finicky chain that used to live in Python scripts. That is when they built out the advanced AI surface: low-code and no-code versions of work that had been reserved for people who could write the scripts, with the brittle connect-A-to-B failures handled under the hood.

The barrier drop is the product. Click a few times, add an agent, add a model, add memory, and it works. What had been historically fiddly became a path from idea to something running. Combined with the marketing shift already in motion, that is the machine behind the eight-month spike: a product that had entered the AI value chain, and a community finally loud enough for the market to notice.

On open source in AI — Meta, Mistral, OpenAI's open-weight release — Oberhauser is warmer than the license debate would suggest. The gift, for him, is a huge army of people who care deeply, are very smart, and often have a lot of time. They do work companies will not, because companies have different incentives and timelines. Most projects go nowhere; those are the ones nobody talks about. The ones that hit something can shape the whole industry.

He pushes back on the cost story. Open source is often not the cheapest path. Hardware sits idle something like 99 percent of the time, and self-hosting can turn out more expensive, not less. What he does buy is the control story. Organizations n8n works with now pick open software mainly for data privacy and data security — they want to know where data lives — and they run it in their own private cloud, not a server under a desk. Saving money is, at best, a secondary reason.

MCP gets a cleaner analogy. Standardization accelerates things even when the standard is not perfect. He has no idea whether MCP will still be the one people use in a few years, but it is a starting point you can build on. "MCP is like the HTTP of AI workflows," he says. It already enables agent-to-agent patterns and the building blocks for marketplaces and plug-and-play automations. n8n's seat, as he draws it, is the orchestration layer between those MCP agents and tools — and, not coincidentally, a strong place to build the tools themselves.

Asked what people actually wire up, Oberhauser repeats that the stack is diverse on purpose. Google's tools are still among the strongest connectors, used by individuals and large organizations alike. Communication apps — Slack, Telegram, and the rest — show up constantly, which makes sense in an agent world: you need a place to talk to the thing you built. Then databases, low-level systems, and a great deal of internal tooling, often the systems that have no off-the-shelf integration. n8n's job is frequently to connect the unconnected.

The user-base explosion of the last eight months made the team nervous. Would quality drop? Would the new people succeed? It did not, which surprised him. The people who stick are either already technical, or they care so deeply about a real problem that they will put the work in. A number of users have started learning to code *because* of n8n: once they saw how much they could do without writing code, they wanted the larger world that coding opens. The community, he says, is full of driven people who help each other.

Being a founder in 2025 feels different. Planning used to be easier; now he lives in a constant state of uncertainty, not fear exactly, in a world that moves every month. The hard part is the tightrope: miss nothing important, and refuse to jump on everything that appears. Signals are messier too, because the open-source-ish world splits into enterprises that pay and a much larger free population that never will. Neither is more important. One helps directly. The other drives all of it.

In practice he leans toward giving more away for free, because what you give away makes the product better for everybody, drives adoption, and drives revenue later, while enterprise features serve only a subset. Right now the job is to capture the market, including the smallest builder, because that is how n8n already gets into the largest organizations. Focus only on enterprise, he says, and you can never come back down. Nobody went from enterprise to owning a space. Google and Microsoft started in the other direction.

Are foundation-model capabilities still compounding at the old rate, or finally flattening? Looking at GPT-5, Oberhauser thinks it feels like a slowdown, which is what you would expect once the early low-hanging fruit is gone. You can throw more compute and more data at the problem for a long time, and then you max out. He treats the pause as temporary. There is so much money in the market, and so many early bets, that some of them will hit the right stage and another acceleration will show up.

That future changes what a developer is. In the past, you asked for X and a developer built X. With n8n and the vibe-coding tools, he sees developers becoming the people who put up the guardrails and empower others to build what they actually need. There is still a role for classical engineers; vibe coders will not build everything. The belief underneath is one n8n has had from the start: the people who have the problems are the best equipped to solve them. Vertical tools will keep doing one job better than any horizontal platform, just as SaaS sprouted a vertical app for everything — which is exactly why something like n8n had to exist, to wire them back together. As vertical AI tools multiply, complexity rises, and you need orchestration again. Either way the market turns, he thinks n8n is sitting in the right seat.

If it all works, the analogy is Excel. Fifteen years ago, people heard "spreadsheet" and thought of Excel. In a few years, when someone thinks "I have to build something with AI," the only name that should come to mind is n8n: the default orchestration layer for building, deploying, and finding agents. The company is already starting to be that default building tool, he says. The user base has been global for a long time — Europe and the US roughly the same size — even though US customers got a worse experience. That gap is why n8n is opening a New York office and hiring hard, from engineers to support to go-to-market, a European company trying to take the US market rather than remaining a regional success.

In the rapid-fire close, the hardest self-truth is that he does not like to say no, and running too many things in parallel has mostly worked, luckily, in a way it would not at most companies. The film he still loves is *Her*, because a few years ago it was science fiction and now it is around the corner. The tool he came to late and now recommends is Granola, simple and very good at meeting notes. The category he would bet on for the next six to twelve months is AI-powered internal tooling. External-facing AI has to be careful. Internally you can take much more risk, and n8n already does. That is the use case he keeps pointing people toward: pick a repetitive internal process, drop in an agent node, pick a model, hang tools on it, and run it.