Brought to you by
Customer.io - Send smarter messages using your product data
Kameleoon - Leading AI experimentation platform
Bolt - Ship AI-powered products 10x faster
Jira Product Discovery - Plan with purpose, ship with confidence
Product Faculty - Get $150 off their #1 AI PM Certification via code AAKASH150
Today's Episode
n8n was the most written-about AI tool in this newsletter last year. I covered it 25 times and ranked it A tier, above Zapier and Make.
Then Claude Code and Cowork arrived, the Google Trends line bent downward, and a tweet went around declaring n8n irrelevant.
So today I went to the person with the most to lose.
Jan Oberhauser is the founder and CEO of n8n, last valued at $5.2 billion after SAP's strategic investment. He says n8n has been declared dead roughly a thousand times, including the week OpenAI shipped its own agent builder, which turned out to be one of his best growth weeks ever.
His argument is not that Claude Code is bad. He uses it. His argument is that they are different products and you need both. Claude Code is where you prototype in ten minutes. n8n is where the thing goes once it has to run every night, survive a model outage, pass an audit, and be handed to someone who did not build it.
So he opened the product and showed me. A live agent build, an eight-minute AI assistant run, an approval gate that stops an agent mid-action, and an execution log that shows every decision the agent made.
Then we got into the growth story. 10x revenue in a year. No lead gen target. No per-seat pricing. A goal of a billion users with fewer than a thousand employees.
10 Key takeaways
1. Claude Code and n8n are different products, and you need both - Claude Code runs Anthropic models in your terminal as an agentic tool. n8n is the orchestration layer connecting your tools, models and data sources on a visual canvas. Jan's own users prototype in Claude Code and migrate to n8n once the workflow has to be reliable.
2. Reliability is fallback models, self-hosting and maintained integrations - Every provider goes down, so you set a fallback model inside the same workflow. You can self-host next to your own data. Each integration is code n8n writes and maintains, so an API change gets fixed once for everyone instead of separately by every builder.
3. Auditability is the thing generated code cannot give you - With code you see the input and the output and nothing in between. n8n replays every past execution step by step, with the data going in and out of each node, and lets you rerun half a workflow from a single data point.
4. AI plus deterministic logic plus human in the loop - An if statement is cheaper, faster and 100% reliable, so it belongs next to the model rather than being replaced by it. Destructive actions sit behind an approval gate, so the agent asks before it sends the email or books the meeting.
5. The AI assistant builds the workflow for you - Jan prompted it the way you would prompt Claude Code. It asked clarifying questions, thought for eight minutes, and returned a working multi-tool agent. Extending it with one-on-one scheduling and Google Contacts took another five.
6. Start small, and skip AI when you do not need it - Companies that try to transform everything at once spend weeks building the wrong thing. Jan's favorite example is a company that automated employee password resets and saved multiple full-time employees a year. The small wins are also the ones that get colleagues interested.
7. n8n wins wherever the work is business-critical - Security orchestration, compliance, employee onboarding and offboarding, DevOps, monitoring. The rule Jan gives is that the more reliability and security matter, the better n8n fits, which covers nearly everything that is not a personal use case.
8. Sprinkling AI on top gets 10 to 30%, being in the value chain gets 10x - Adding an AI button somewhere is not a strategy. n8n's bet was that when someone decides to build an agent, they build it in n8n. That choice is what produced 10x growth in a year.
9. Deleting the metrics that would make money faster - No lead gen target and no per-seat pricing, because profitability means n8n can think in years instead of quarters. The company does not push free self-hosted users onto paid hosting either. The internal goal moved from a billion in ARR to a billion users so nobody confuses the mission with money.
10. Talent density over headcount - The target is a billion users with fewer than a thousand employees, which keeps the hiring bar high. Jan wants tinkerers who run home automation and understand scale, evals and reliability. Live problem-solving beats take-home tasks now that AI can do the take-home for you.
Go Deeper
Build your first n8n workflow with Pawel Huryn’s guide.
Then follow Mahesh Yadav’s learning path in How to Become a Builder PM. Prototype your next agent there, and promote it into n8n through the MCP server.
Write 20 test cases and run them as an eval.
***
You can get my podcasts on Apple, Spotify, and YouTube.
***
If you want to advertise, please email productgrowthppp[at]gmail[dot]com.