It turns out that react's programming model is great at constructing and executing DAGs. The resulting code is much simpler, and more expressive.
And the React's focus of composition over abstraction is 100% what you need in the LLM world. It enables sharing code across large teams by default.
When you have serializable props, outputs, and pure functions (properties of both the React and GenSX component model) then it is trivial to layer things like durable execution on top of it. Might give you an idea of what is soon to come!
I've never been bullish on no-code. Every couple of years there is a new unicorn darling, but none of them ever reach escape veclocity. They just seem to replace each other. An unfortunate category of product where the end state is just graduation to either learning to code or hiring an engineering team.
Happy to discuss "no framework", our experiences building long-running agents that make thousands of LLM calls, and why we think that the infrastructure that runs the internet is going to look very different in the next few years.
My perspective is that the goal of JSX isn't to make XML "elegant" but to offer a way to dynamically build more complex trees and even DAGs programmatically. Best of both worlds between declarative and easy to read, and expressive via dropping into plain JS.
(I'm Evan - the founder of GenSX, and always open to feedback if you have any!)
Totally understand that JSX and React-like frameworks aren't for everyone.
I have worked on several DAG frameworks (such as Pulumi), and JSX turns out to be a happy medium between declarative and easy to read, and dynamic allowing you to drop into plain JS, use for loops, if/else, etc. GenSX components are just plain functions and can be called as such when it makes sense!
React's component model is well understood by developers, and easy to build tooling on top of. Components are reusable by default since they're functional, which opens the door to an OSS community to share useful building blocks.
Full discloser, I'm Evan (the founder) and welcome any feedback you have as we're building. Sorry for the late response, I didn't notice this got posted.
We have an equivalent of Atlantis called Pulumi Deployments[1]. The benefit of the Deployments platform is that it is entirely API driven. In addition to defining CI/CD and deployment in configuration/code, we offer APIs that let you do this programmatically. Great for platform automation where you are setting up hundreds or thousands of stacks.
In addition to `git push` workflows, we also support other deployment triggers such as a REST API. This is pretty unique, and let's you do things like build RESTful infrastructure APIs [2] ontop of the deployments platform.
If you're looking for a search backend without having to worry about managing infrastructure, you should check out Azure Search. I'm a dev on the product. If you're interested in trying it out let me know. You can ping me at my username @microsoft.com if you have questions or need pointers.
Hi, I'm Evan an Engineer on Azure Search. I'm not aware of any published in depth comparison of the two services. We'd love to hear your thoughts and feedback if you'd like to give the service a try. We have a free tier that allows you to load up to 10k documents. If you do give it a try, please feel free to reach out to me at username at company.com.
It turns out that react's programming model is great at constructing and executing DAGs. The resulting code is much simpler, and more expressive.
And the React's focus of composition over abstraction is 100% what you need in the LLM world. It enables sharing code across large teams by default.
When you have serializable props, outputs, and pure functions (properties of both the React and GenSX component model) then it is trivial to layer things like durable execution on top of it. Might give you an idea of what is soon to come!