@Shabarish_B@galligator Congrats on the App Store approval! GLP-1 space is interesting right now. What's been the hardest part so far getting it approved or figuring out how to get first users?
@jardim_weber@boltdotnew Security is one thing, but most first-time app builders I talk to get blindsided by something else after launch, App Store discoverability, IAP setup, or just zero downloads despite a working product. What's your app about? Curious what you're seeing early on
I’ll admit it: I’m slightly addicted to AI.
Comparing products before buying something? AI.
Trying to structure a workout? AI.
Figuring out what to cook? AI.
It’s genuinely useful for all of that.
But there’s one thing even the strongest models still seem terrible at.
Advice.
Recently, I asked AI to help me make a decision.
The response went more or less like this:
“Choose option A.”
“But A has these downsides…”
“Then go with B.”
“But B has problems too…”
“…in that case, probably A.”
It turned into a literal endless ping-pong. The AI was openly admitting it was changing its mind, and then continued changing it anyway.
You can try to patch this with better prompts, stricter instructions, even system-level scaffolding.
It doesn’t really solve the problem.
AI is great at expanding your perspective. It can surface considerations you missed, organize messy information, and give you a better map of the situation.
But when there are no clean technical parameters, and the real challenge is judgment, conviction, or trade-offs under uncertainty, it still falls apart surprisingly fast.
That’s worth keeping in mind before handing AI too much of the steering wheel.
Have you had a similar moment with it?
What interests me is not AI added as a chatbot somewhere in the corner.
Honestly, that version never felt that useful to me.
What feels like a real shift is MCP-style integration.
When AI can actually operate the tool, move through the interface, make changes in the project, and turn intent into action, that starts to matter.
A simple photo edit is already enough. Color correction, small adjustments, getting an image to look the way I want none of that feels natural to me. But if I could just tell Photoshop what I want and let AI handle the messy part, that would be a game changer.
You can already see early versions of this direction in tools like Lightroom, where Adobe has been expanding AI-powered edit workflows and AI edit management across its products.
That’s the kind of AI future I’m waiting for.
Even in FlutterFlow, someone from the community has already created an MCP package, so you can use tools like Claude to operate directly inside FlutterFlow.
In 2026, building a mobile app in weeks is no longer something unusual.
Not every app, obviously. And not every fast build is a good idea.
But getting from idea to something real, usable, and in people’s hands now takes much less time than most people still assume.
I talked about this in my latest episode:
https://t.co/MKa1bODL4g
Built a social media app in low-code recently.
Honestly, I didn’t run into the kind of limitations people usually expect when they hear “low-code”.
Feed, interactions, logic, product flow - all there.
At that stage, the bigger challenge wasn’t the tool. It was making good product decisions and keeping the scope under control.
That’s usually the part people underestimate.
I recorded the full breakdown here for anyone curious what that looked like in practice:
https://t.co/1405Z7xwag
Small detail I appreciate in early-stage apps:
In-app contact that doesn’t push you to email.
One tap. One message field. No account portal. No 10-field form.
Most users won’t open their mail app to report a bug. That tiny friction kills feedback loops.
Even better?
Attach app version, OS, device, build number and logs automatically.
User writes: “It crashed.”
You already have context.
In the first months of a product, speed of understanding beats adding another feature.
Vibe coding works - until it doesn’t.
I tested it on a constraint-heavy project that wasn’t a clone of something that already exists.
Four hours later: nothing usable.
Opus 4.6 was extremely productive.
Clean abstractions. Lots of files. Constant restructuring.
When something didn’t work, it added more code.
What it never did was question the direction.
It kept optimizing for something that compiles - even if the premise was wrong.
AI performs best when patterns already exist.
With real novelty, confidence stays high while reliability drops.
The screen fills with code. It looks like progress.
Sometimes it’s just acceleration in the wrong direction.
I’ve spent years building mobile apps the “proper” way.
Long timelines.
Custom everything.
Perfect architecture.
Recently I started focusing on something else:
Helping founders ship MVPs in weeks, not months.
Speed changes everything.