“It can literally kill your company”: n8n’s case for model-agnostic AI

Asked where his company sits in the AI stack, Jan Oberhauser reached for a car. The models, the n8n founder told a fireside audience at the Raise Summit in Paris this week, are the engine. Everything else, the vehicle, the streets, the rules that let the engine actually take you somewhere, is the part he is trying to build. It is an unfashionable place to stand in a year when the engine gets all the attention, and that is rather the point.

n8n, founded by Oberhauser in 2019, describes itself as the orchestration layer for AI, or, in a phrase Oberhauser has used elsewhere, the “Excel of AI.”

The pitch is that the platform sits between the models and everything a company already runs, stitching large language models, deterministic code, and human sign-off into workflows that survive contact with production. What it deliberately does not do is make a model of its own.

That model-agnosticism is not a recent marketing turn. It is baked into how the company licenses its software. n8n runs on a Sustainable Use License, the “fair-code” model the company coined and adopted in 2022, having moved off Apache 2.0 with a Commons Clause. The source is open and self-hostable, with commercial use restricted.

The design goal, Oberhauser said, was that a user should be able to own their data, self-host, and avoid lock-in, no matter what happens in the outside world. For large organisations, he argued, that freedom is the whole appeal.

The past 12 months turned that argument from theory into demand. As OpenAI, Anthropic, and DeepSeek traded releases, companies rushed to wire models into their operations, and a platform that could plug any of them into anything found itself in the middle of the boom.

n8n now reports more than 1,400 enterprise customers and around 1.7 million monthly active builders, with Meta, Vodafone, and Mercedes-Benz among the names on its roster.

“Prepared to switch, not switching”

The interesting nuance, on Oberhauser’s telling, is that most customers do not actually swap models very often. They want the option to. Every switch breaks something, he noted, and a model from a different vendor, or a self-hosted one, or a new version of the same one, behaves differently enough that you need evaluations in place before you can trust the change.

The reasons to move are real, though: cost at scale, latency, quality, or the chance to fine-tune an open-source model and run it cheaper and faster at once. Most firms, he said, are building the infrastructure to be ready rather than migrating today.

What has accelerated those conversations is sovereignty. Oberhauser said boards are now asking what happens if a provider raises prices, gets acquired, or loses access in a given geography, the kind of dependency that, as he put it, could “literally kill your company.”

The rise of shared standards for connecting models to tools has made switching look less like a research project and more like a requirement, and European customers in particular are pressing on where their data and their models live.

Being able to move “reasonably fast” from one provider to an open-source alternative, he suggested, is now a form of insurance.

That framing is doing real commercial work. In May 2026, SAP invested in n8n at a reported 5.2B valuation and agreed to embed its visual workflow canvas inside Joule Studio, SAP’s agent-building environment.

Mercedes-Benz, meanwhile, has rolled out n8n as a global automation platform, drawn in part by the self-hosted, cloud-agnostic deployment that keeps sensitive data on its own infrastructure.

On crossing from bottom-up adoption to enterprise sales, Oberhauser described a patient approach: give the product away widely, let workloads pile up, and wait for the moment a large organisation needs single sign-on, enterprise support, and the rest, then have the conversation. The harder message he wants to land is about return.

He pointed to research finding that only about 5% of corporate AI efforts deliver real value, an echo of MIT’s much-cited “GenAI Divide” report, which found 95% of enterprise pilots produced no measurable profit-and-loss impact.

The winners, in his account, treat AI as one component rather than the whole system, pairing it with deterministic logic that is fast, cheap, and reliable, plus a human check at the end. It is a distinctly unglamorous prescription.

In a market still intoxicated by the engine, the company betting on the car, the streets, and the rules is wagering that the unglamorous part is where the money finally shows up.