The embedding bill: an early n8n lesson I now apply everywhere
Early in my career I wired an expensive third-party embedding service into a client’s n8n workflow without checking its pricing — and the client paid for my assumption. What changed in how I work after that.

On one of my first automation projects, I connected an n8n workflow to a third-party embedding service because it was well-known and easy to wire up. What I did not do was read its pricing page properly — or model what our actual usage would cost. The workflow worked beautifully. The bill did not.
The client took a loss on my assumption, and I owned my part of it: I migrated the workflow to a cheaper alternative immediately, and the project recovered. But the lesson cost more than the invoice, and it should have.
Since that day, pricing is part of my design step, not an afterthought. Before any third-party service goes into a system, I check its price page, its rate limits and its free tier, and I estimate the cost at the usage the system will really see. Tool-calling agents and embedding pipelines can spend money while nobody is watching — the design has to include the meter.
It is the same habit as checking a library’s licence or a database’s limits before building on it: verify the constraints first, because "it works" is only half of "it is fit to ship".