Hot Take: Low Code/No Code platforms die as LLMs get better | Hacker News Reader

Hot Take: Low Code/No Code platforms die as LLMs get better

12 pointslivewirecrazy19 comments

The better LLM‘s get at generating code the less of a need there is for low code platforms.

More and more articles come out every day about how you want be on the right side of LLM’s as they improve.

Low code and no code platforms have an appeal of no source code - but having no source code makes it very difficult to train an LLM to control it.

Extending this thought it's possible to rationalize that building with any form of proprietary thing is going to be harder and harder the better LLM's get. You want build using something that has infinite training data on the web that the largest models are already getting trained on.

Software platforms in many industries are also known as having a winner take all set up -- the whole network and ecosystem effect. LLM's extend this to software stacks for builders.

View on HN

We are living in the "low-code" of the 1990s, who in turn...

In 2015, I thought React was low code. It came along and eliminated the need of dom update functions. Imagine all the Jquery or Vanilla functions to sync the data/dom for "add to cart", "increment item", "clear cart" across the product display and top bar and all the subtle out of sync bugs.

Looking at React ecosystem over the years, I felt the pain of the people who in 1954 believed the authoritative "Since FORTRAN should virtually eliminate coding and debugging...".

https://www.softwarepreservation.org/projects/FORTRAN/Backus...

There was never a need for Low Code platforms to begin with, unless demand from wishful thinking is included too.

To accomplish the same thing in regular software languages and low code platform takes the exact same learning curve. If the low code platform is good, which they are often half arsed. And learning a regular software language is much more useful down the line too.

Fully agree. As I end up pointing out too many times in my day job - a problem has a minimum level of complexity that you can't go below and still solve the problem. Using a low code platform still requires you to do all the hard work of figuring out a solution, but you then have to express in the confines of your low code solution.

One of my co-workers is trying to use one of Microsoft's low/no code platforms to do something with CRM and it's taking her months. I could've built something with code in significantly less time.

As far as I'm concerned, these platforms are borderline snake oil.

In the end, I don't see the distinction. I expect that instead of an LLM writing text, it may as well make the final thing and you can 'view source' if you like or just use it. So the LLM is the low-code platform.

What will be interesting is what chosen languages/frameworks are the most effective in LLM-based no/low-code platforms. I've wanted forever for developers to choose/use better languages and tools but that's always too varied to execute at a scale beyond early startup. I could see the low-code AIs/LLMs having an advantage using better communication/action interfaces to make apps. Geez this could all happen fast.

The top low code platform in the world is Excel. I don't think LLMs will replace it any time soon.

To some degree it's true unless nocode/lowcode platforms will adapt to the situation - I used thing called Xano for a bit recently and will use them as the example of very poor adaptation (so far). Their insistence on being "no code" as much as possible became really annoying and tedious after a while - some examples:

* through multiple iterations of back and forth with LLM assistant I can get DB schema but there is no way to import whole schema with relations, indexes, views described with SQL statements into that nocode app - extremely annoing to have "manually", tediously click and recreate the thing

* same with their drag&drop "functions" stacks builder - very annoying to fiddle around and click click click when in the back of your head you have "damn, llm could generate that block of code in 1m and be done"

* don't even get me started with their custom DSL "expressions" - yet another thing not usable anywhere outside of their platform, despite potential and usefulness I have absolutely no incentives to try to learn that

At this point I really think that I would be faster with code and LLM than this nocode DBaaS

> Low code and no code platforms have an appeal of no source code - but having no source code makes it very difficult to train an LLM to control it.

Low/no code tools don't need to be like that. The next generation will be the ones beating the current incubents. As LLM can generate code, LLM-powered no-code tools will deliver high level products with a fraction of the effort needed nowadays. OpenAI, Anthropic and the likes won't be capable of creating the best UX for consumer apps using LLMs. That's a great opportunity there.

I am founder of DronaHQ - low code platform. Interestingly Today we launched AI capability to build internal tools with AI in beta. The idea here is to marry the power of LLMs and underlying low code platform to get to the end state even faster.

But I do see the merit in this thinking that LLMs can eat LCNC platforms for breakfast. However I am excited to see how with time a near perfect applications can be churned out with about 80-90% work done and engineers can figure the last few miles and reach the finish line but 50x faster.

I thought about that in the same way and I think you could be right, as prompting your software could be even more barrier free than using low/no code tools. On the other hand it is not to be over estimated that an LLM doesn't need that much training data for good results especially in a specific niche. What I want to say is that if LLMs could enhance low/no code plattforms it could be even more powerful than using the currently most-used programming languages and frameworks

The low-code, no-code niche that I see is writing shaders using visual graphs - blender, unity, unreal etc. I don't see that being replaced by LLMs in the near term. The context where you're expressing what you want is represented quite nicely by the graph. If you're saying to an LLM make me an animated, iridescent, force field like shader seems a way off to be able to get to what you really want.

In fact, the lowcode/nocode platforms produce very well structured code behind the scene and they can train LLMs to build apps following their well defined structure, which means their apps can still take advantage of the LLMs but will have less bugs than just letting the LLMs produce a stand-alone app. Of course, the LLMs can produce more versatile apps than lowcode/nocode platforms but I don't see the LLMs as the competitors.

I believe Low code / No code platforms evolve to use LLMs to actually still abstract out all the code bits but become even way more powerful. It’s cool to use cursor with sonnet but it’s still requires a level of being comfortable with writing code, local builds etc.

No code platforms can easily abstract this out one more level. The constrained outputs of these platforms are still sufficiently high enough to build lots of varied use cases.

Agree with this take (disclaimer: working on create.xyz, which is step in this direction) It abstracts more of the things that trip non-devs up (deployment, etc.) but retains the things that makes code powerful (customization, etc.)

Gdevelop might be described as a low code platform as it's language is visual (like Scratch on acid). It already has "Generate a game with AI" built in. It's not very good.

What makes you think LLMs will get better? The more we've seen AI develop the more fine tuned everyone's senses have got hyper-sensitised to what's AI and it just reeks of the uncanny, rendering it ultimately useless.

I can only project from my own experience, and I think that no/low-code fail because every platform I tried (and I tried many) still fail hard to become RAD.

They aren’t getting eaten by LLMs, they just suck and that’s already a low bar to die due to a reason de jour. Cool looking toys to experiment with in a free money economy.

I agree with your statement. Also, the recent interview with the former Google CEO revealed many things about what AI will be capable of in the upcoming years. If you think about it, it's insane!

Can't they directly add API calling to LLM in their platform?

Pre-sentience = wild wild west post- sentience = every sci-fi movie/book you've ever seen/read.. In the meantime every possible scenario being drummed up is mere fear-mongering because the post event will radically change the dynamics of the pre event. We simply have ZERO precedence in human history to guide our speculations. And the power...the intent...the results...cannot by the best of means begin to paint a picture with any degree of accuracy. Sit back and enjoy the ride is all I can say. Oh...and make nice with our new masters while in pre event somehow if you can. Cross your fingers it will have any impact in post.

Reply on news.ycombinator.com