As a non-coder I was obsessed with automation tools like Zapier when they first came out. After that, I tried very hard with n8n but just found it impossible to work with until I read some post by some developer that said "just do it in Python". Took me a while to really take that in given I can't code but then I got Claude to write some Python scripts automate some work spreadsheet and API work and I haven't looked back since.
I think something like this could have worked several years ago but now it's overkill.
The thing in TFA looks like it can draw trees.
Can it do DAGs? Can it do directed graphs with loops?
This is, to me, what sinks most of those services like Zapier and n8n in power-user market, and dooms them to become marketing platform for third-party integrations, concealed by some drag&drop complexity. Because the moment you want to do anything that isn't a direct ETL job with SaaS for databases, you discover the thing is trying very hard to not be Turing complete and there's basically no easy way to express any logic without shelling out to code, if that option is even available...
A product I wish existed is basically something like Zapier but with code. That is, webshit SaaS with quick IDE and (that's the important part) full auto-complete of all active integrations as function calls. Plus some basic correctness checking.
Rationale:
It's easy to write the equivalent of a low-code flow in code; almost trivial if you can tab-complete a dropdown with icons and descriptions for each symbol:
But in real coding system alternative to that, you'll also need to write 10-50x more code to deal with imports, build system, setting up APIs, constructing objects, and dealing with other incidental bullshit. But take away that, add some fancy autocomplete, and allow for control flow with loops, and I think you'd get all the benefits of low-code tools + all the benefits of coding + as a bonus, you'll make it easy for agents, too.TL;DR: the main value of low/no code tools isn't in the drag&drop UX - it's that they regularize the APIs of the integrations, solve the discovery problem of those APIs, and hide all the incidental bullshit behind the scenes. In principle, there's no reason you couldn't just do the valuable parts, wrap it in an API, and let users run code against it on the platform.
Try windmill.dev