The examples they show are someone typing something into a prompt and then some agent starts coding.
But the hard part in automation these days is not writing software. It is "how does the software get the credentials to access data and services? What APIs does it use?". So before being able to type a prompt and then use the software created, there must be the hard part: Setting up all those interconnectivity pieces. Making sure the software can access and change data, trigger actions, but only within a certain scope.
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.
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.
thb im just tured of every new tool claiming they change everything. like does this actually work or is it just another wrapper that will break next week.
> We’ll model it in Ballet on your real systems in a 30-minute working session
You drastically underestimate how bad most systems are. All of the vendors in your examples have good APIs. Those vendors already did the hard part.
All of the systems we need solutions for are moving targets that are breaking constantly. You can’t model those in 30 minutes because the ways in which they break don’t show up until months later.
Absolutely. And scale does not matter. I had issues with top tier systems like docusign, tableau etc in beforeGPT era. Many undocumented issues and when you write support usually they only were pushing to buy top tier support package, but never fix anything.
Yet another automation? How do you plan to win the market when you add just 1% of extra value? In a few weeks it's commodity. The fact that it's easy to build nowadays don't mean you need to build every cool idea that pops up your mind...
However, I might be completely wrong - and in that case, I wish you luck and success on your journey!
"Built by the team behind Brainfish"
Perhaps I'm the wrong market for this project (despite much of my bread and butter being types of workflows) because I have no idea what Brainfish is.
I don't get what this is about.
The examples they show are someone typing something into a prompt and then some agent starts coding.
But the hard part in automation these days is not writing software. It is "how does the software get the credentials to access data and services? What APIs does it use?". So before being able to type a prompt and then use the software created, there must be the hard part: Setting up all those interconnectivity pieces. Making sure the software can access and change data, trigger actions, but only within a certain scope.
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
thb im just tured of every new tool claiming they change everything. like does this actually work or is it just another wrapper that will break next week.
> We’ll model it in Ballet on your real systems in a 30-minute working session
You drastically underestimate how bad most systems are. All of the vendors in your examples have good APIs. Those vendors already did the hard part.
All of the systems we need solutions for are moving targets that are breaking constantly. You can’t model those in 30 minutes because the ways in which they break don’t show up until months later.
Absolutely. And scale does not matter. I had issues with top tier systems like docusign, tableau etc in beforeGPT era. Many undocumented issues and when you write support usually they only were pushing to buy top tier support package, but never fix anything.
- zapier
- n8n
- tonnes of these keep popping up
Yet another automation? How do you plan to win the market when you add just 1% of extra value? In a few weeks it's commodity. The fact that it's easy to build nowadays don't mean you need to build every cool idea that pops up your mind... However, I might be completely wrong - and in that case, I wish you luck and success on your journey!