Hi Weweb team, absolutely glad to be back from leave and just read @Raphael update. Now no flowery words let’s get to it:
I love your product. Weweb is the best and most robust no-code frontend tool in the world as far as I know it, and trust me I know a bit about no-code tools (I was using Graphite GTC, Blockly & MIT app inventor long before No-code became cool), so I consider myself a bit knowledgeable and in my experience nothing comes close to weweb’s flexibility and extensibility for web developement. But you already know this. So to my other feedback
Weweb AI, I have to say without equivocation (and very respectfully) is awful, I personally do not think it adds much value beyond very basic use-cases and help; and for something that the team clearly put a lot of work into and spent a lot of personal capital on, it needs reworking. I ignore it mostly (I have never used it since it moved into that side bar and was not in the editor pane, I just ignore it) I am a coder, I code components, so it’s not an issue. But recently I decided, I wanted to actually see what it could do, and it was really just not good.
It doesn’t seem to follow instructions well - even simple ones, once you have more than one if/else situations it is definitely going to fail
The context window seems TINY (not even sure which model powers it but boy…)
It is TERRIBLE when working with coded custom components (again, here I think it is the context window killing it). It is just really bad and I care about this product so I had to come say it here, cos there’s no private place to go unless I open a ticket or something.
What would I hope weweb did differently?
Narrow the use case. Right now it is trying to be a good agent for everything - help you edit UI elements, help edit workflows, help edit coded components. It’s too much for it. And it is not great at any of it. Very optimistically, it has encouraging performance when working with basic UI elements meaning it can start there as it’s lever). Right now, it is NOT a product retainer, I daresay I do not think many people who are here are hooked because of Weweb AI. And for something with great potential, that’s not cool. My point here is, could the use of the AI perhaps be narrowed to a single use case, and then all effort made to ensure it is truly outstanding at that single use case. For example, say you say the use case is generating UI, now do all of the work to make sure it generates great helpful UI even under very tough conditions (and wedge there, then grow it). Trying to slap AI everywhere and harming the editor in the process and the AI ending up being unremarkable, is just not a “winning“ strategy in my eyes, even if I want you to win no matter what.
The context window. That AI suffers from context rot, I have absolutely no doubt about that, the context window is not being well managed perhaps the team could check tools like NIA from Nozomio Labs (they’re YC backed too I think). Or look at how to better handle the context window cos right now, today, unless we are being encouraged to lie to you, the unfiltered unadulterated truth about the AI is that, it’s not competent. Simple.
You have my admiration always for this wonderful platform. Thanks. Best wishes.
Thanks for the great feedback and totally agree. We are completely revamping our agentic AI architecture. We’ve been at it since last November and hope to release a new AI in March/April of this year.
The result should be an AI that is much more helpful, much more granular and makes very few mistakes thanks to better context management. We will also improve the UX around it: being able to plan, to stop the AI while it’s building, to easily revert, to tag specific elements/workflows/functions/variables, etc.
No one cracked yet this AI-to-JSON-to-App architecture for nocode builders. It’s actually a very complexe technical problem to solve for an application of WeWeb’s scope, but we are getting there.
Once cracked, users will enjoy a reliable AI-powered development mixed with the full control of visual programming and JSON guardrailing (meaning the AI will only be able build within pre-set, deterministic guardrails, solving the #1 pb with vibe coding right now!).
In our opinion, this is the grail for nocode right now and will have tremendous value for users building real, production apps. So, I am both excited and nervous, because this AI has an incredible potential and the problem is incredibly hard to solve. It takes a lot of time, but we are getting there.
That new version in March will be a beta version of the new agentic architecture, but it should already deliver much better results compared to what we have in production right now.
WeWeb is one of the best visual frontend builders on the market, and that’s exactly why the current AI push feels so misaligned.
I’d actually love a real AI for no-coders — but right now it’s not a “no-code assistant”. It generates Custom JavaScript instead of native actions, doesn’t understand WeWeb components properly, doesn’t verify if the result works, and produces tons of extra variables/complexity. So it often makes the product harder, not easier.
Strategically I’m also confused about the audience: vibe-coders already went to Lovable, and WeWeb has always been a pro-builder tool (visual editor + low-code power). It won’t become a pure no-code leader anyway — JSON and workflow exception handling are part of WeWeb’s DNA.
If I remember correctly, the old public roadmap and feature voting were pointing elsewhere (including PWA → native). From a market perspective the clearest battle WeWeb can actually win is FlutterFlow — not AI-first tools.
I really hope the team narrows the scope, stops trying to cover everything with the same team, and doubles down on what already makes WeWeb win.
True indeed x2. To build something valuable you need to know how things work. And if you know that, Weweb provides you with the most flexible and productive building experience I ever knew.
It does require some knowledge, but that’s not a flaw. It’s the border between the whim and the idea:
low effort vibe-coding whim is a shorthanded path with tremendous technical debt,
demanding knowledge is a way to make an idea real. And, concurrently (if you allow me to be little poetic :), becoming the better version of youself in terms of logic, coding, people understanding, etc.
@Raphael , we all understand the willing to turn Weweb into the most easy-to-use-and-build-complex-things machine, because that can really increase the audience.
But wouldn’t that be the short-term increasity, which will cost you a huge amount of effort? Wouldn’t unhappy users leave Weweb after they fail due to incompetence in prompt formulating?
Meaningful vibe-coding that generates a meaningful result still requires from you to know how the desired product logic should work. Even with unmistakable AI an ordinary user could stuck in an endless loop of correction something that works properly, but the user just doesn’t see that
Long story short: going towards AI is good and justified. We just have doubts about AI being the most important thing in Weweb development.
If not for AI, proper governance at workspace level, improved performances and branching, what would you like us to focus on in 2026?
If the answer is SSR and better support for PWAs, it is great because it will come at some point. We just need to properly deliver on these other items first (which will make the largest majority of WeWeb users happy).
I have a very critical view of the trendy “know all do all” AI approach.
“More AI” does not answer the question of how AI creates a meaningful impact within the app. I think a more valuable approach would be to build smaller AI features that are isolated and robust, instead of going for an all-out, AI-magic approach.
Another interesting approach could be to use AI to educate and empower users in a way that builds knowledge and actually helps them, rather than “throw an idea at an AI and let it do some magic.”
It would be great if I could ask the AI whether the workflow I just implemented has any potential risks, whether it is cleanly built, or if there are better or more robust ways to approach it.
The philosophy right now seems to be “users ask for it, so we should deliver it.” Don’t get me wrong - this is great and sends a strong signal that WeWeb is evolving with user interests in mind. But it does not necessarily validate the entire business case.
Are the users who want (more) AI functionality also relevant revenue drivers for WeWeb, or are they low-usage, low-monetization customers? Is implementing AI at this scope really justified, or is this outlook too short-sighted?
