Looks like this

I, too, read "still pooping"

🤯

Seems a bit sus compared to fable and sol tbh. Idk who is going to pay for grok bots if it doesn’t even go with x sub. There is no point in even trying it for now

have you tried Ploy? it really worked for us , the results are amazing

lgtm

Grok avoiding false security sermons is useful. Still trusting Codex when the landing page needs real code, though.

Looks like what? Need the actual setup before I can tell you if it works.

Yes, we know how to access a website

The smartest AI products should probably be model-agnostic from day one. Today's #1 model can become tomorrow's expensive legacy dependency.

Reminds me of what @polsia is doing. At some point you could have an agent build every idea

It sure is

Where are the ads?

What do you mean?

Why not sell ads for developer services? I’d think that’s very valuable space for a lot of companies.

Clicked, and now I'm curious what this is 👀 always down to check out something new you're working on

looks pretty fast too.

about to be lit ! 🧠🌿 Cheers @levelsio

I don’t think I will ever get tired of Peter’s ideas xD. You are a genius xD. Question (prob stupid): how do you earn money on ?

😂😂😭

is being prompted to use a levels-like stack? i have been encouraging my agent to try this build with no external dependencies

Check the attached image before replying. If it is the video editor agent, the state handling question from last week is still the sharpest thing to ask.

'Without the constant preaching and blocking you from doing anything due to SeCuRiTy' is going to resonate with more builders than any benchmark score would.

The real test isn’t the first landing page—it’s whether Grok can preserve style and conversion logic across 100 radically different ideas. Speed becomes meaningful when the output remains editable.

🤨🤨

With deepseek 4 pro should be extremely cheap and fast

The refusal tax is real and almost nobody prices it into model choice. A model that benchmarks 5% smarter but lectures or blocks on a chunk of requests loses on throughput for autonomous work. For agent pipelines we now weight refusal rate as heavily as capability.

How are you finding it? It’s only recently been released. Anecdotally better or equal to Fable?

retired specialists is a very good test case lol

To get a quick landing page looks ok but you need to personalize the shit out of it

This is how model competition should work. Switch a live product, measure output, keep what performs. I mostly use ChatGPT and sometimes Grok. Benchmarks are theatre until the model survives customer-facing work.

Its not great