This is super cool but I feel like I could just build this myself with an LLM agent, so I get worried about the value proposition for services that try to charge for this. I guess really what you're paying for is some time saving and so that you don't have to burn your own tokens to build it?
As far as I can tell, this isn’t a product, and the page has a prompt to have your own coding agent generate these. Stepshots is a bigger product that offers interactive demos, which the page contrasts with these.
I absolutely love how LLM are incredible tools for this kind of applications.
And yet, everytime I see this, to me it's clear how the labs are doomed.
These are the use cases we see, and kinda feel they're worth the hype.
But in the end it's all superfluous add-ons we wouldn't have bothered with in the past and be fine anyway. The reason we had never done something like this it's because the cost/benefit ratio was off.
Here is where I think the entire financial contraption falls apart.
Because the moment the cost of doing this stuff becomes again financially comparable to the mental/resources cost we needed to program it manually (and therefore refrain from doing it), we will start just... not doing it anymore.
If we instead will be able to still do it, it will mean that the cost of this has stayed as low as now (or gotten to free due to open models), and that will spell doom for the financials of the services enabling the LLMs to do this.
It's a catch22 for Anthropic and OpenAI, and either way they're dead in the water (financially, again, because their products as tools are indeed incredible. yet, they don't create enough value).
This, and the fact that other really value-adding breakthroughs that would justify the trillion dollar bet they're still making are nowhere to be found, except in the doomsday predictions of the AI-psychotic.
Here is where I think the entire financial contraption falls apart. Because the moment the cost of doing this stuff becomes again financially comparable to the mental/resources cost we needed to program it manually (and therefore refrain from doing it), we will start just... not doing it anymore.
If we instead will be able to still do it, it will mean that the cost of this has stayed as low as now (or gotten to free due to open models), and that will spell doom for the financials of the services enabling the LLMs to do this.
It's a catch22 for Anthropic and OpenAI, and either way they're dead in the water (financially, again, because their products as tools are indeed incredible. yet, they don't create enough value).
This, and the fact that other really value-adding breakthroughs that would justify the trillion dollar bet they're still making are nowhere to be found, except in the doomsday predictions of the AI-psychotic.