You can falsify it through deduction, thinking of all of the situations the chain of events cannot lead to. Over time, with enough conclusions, you can focus into the remaining plausible directions. This is similar to the game of 50 questions.

At work, I was involved in a project where a large number of individual tasks defined as declarative code had to be translated into JS based equivalents. Due to the unpredictability of each task we would have to do this pretty much manually, one by one. I would estimate at minimum 2 months of grunt work for 4 entry level engineers. Thanks to coding agents and LLMs we were able to achieve this task in a week. Quality of the end result is top notch.

If that's not a product ... then I don't know what it is.

- What was the state of AI/LLMs 5 years ago compared to now? There was nothing.

- What is the current state of AI/LLMs? I can already achieve the above.

- What will that look like 5 years down the road?

I you haven't experienced first-hand a specific task before and after AI/LLMs, I think its indeed difficult to get insight into that last question. Keep in mind that progress is probably exponential, not linear.

you don't see the products because not all AI-assisted dev products are AI wrappers. These products look like regular software, both internal company tools and external customer facing ones.

There are people all over the place building stuff that would've either never been built, or would've required a paid dev++.

I built a whole webshop with an internal CRM/admin panel to manage ~150 products. I built a middleware connecting our webshop to our legacy ERP system, smth that would be normally done by another software company.

I built a program with a UI that makes it super easy for us to generate ZPL code and print labels using 4 different label printers automatically with a simple interface, managed by an RPi.

I have built custom personal portfolio websites for friends with Gemini 3 in hours for free, smth that again would've cost money for dev or some crappy WP/Squarespace templates.

As the other user said, the progress/changes are not distributed evenly, and are impossible to quantify.

But to me whose main job is not programming (but who knows how to code) but running a nom-software business, the productivity gains are very obvious, as is the fact that because of LLMs I have robbed developers of potential work.

the world does not need more shitware. We need medical advances, scientific breakthroughs and societal shift to improve wellbeing of all people. these things are much harderthan writing shitty sofware and we will need not the current AGIs(Goggle Gemini 3 Pro and ChatGPT 5.2 Thinking) but ASI to solve them.