Without a usable model trained on lots of data, it would be hard to give a demo.

it's not beyond the realm of possibility to train it. I emailed 30-40 people already with 37 screenshots including text generation and training. Not a single response. I thought for the time being this path is preferable over huggingspace/github. I'm not ruling out just open source and large .bin pretrained file...i'd need to think about it more.

To add to that, you can always make your drone/bot loiter for a while before getting a fix.

Vereenigde Oostindische Compagnie (VOC), or Dutch East India Company

The frontier of how good models are also shifts and will remain ahead of local models unless we hit some dead end limitation in the algorithms themselves. A ceiling so to speak on how good LLM can get before the law of diminishing returns starts to apply.

i don’t understand. all models are local models, they’re just not running on your machine.

Is the discussion supposed to go back into some training set? Or this is only for humor?

I think it would be the opposite and we are all in for a rude awakening. If you have tried playing with Opus 4.6 you would know what I am talking about.

I tell my colleagues we're in the instantaneous peak of the AI developer relationship, especially for code monkeys. We're still valued, still paid really well, and our jobs will get easier and easier probably for the next 5-10 years! After that, maybe not so great for many of us, with the developers that use software as a means of their actual profession continuing to do just fine (hard math/science/optimization/business planning/project planners/etc).

I think it's going to be an amazing shift from those that know intricate details of software to enabling those that have the best ideas that can be implemented with software (a shift from tool makers to tool users).

I think many developers misunderstand the quality of software that people outside of software are willing to live with, if it does exactly what they want when they need it. For a user, it's all black box "do something I want or not" regardless of what's under the hood. Mostly "academic", things like "elegant" and "clean" and "maintainable" almost never practically matter for most practical solutions to actual problems. This is something I learned far too late in my professional career, where the lazy dev with shite code would get the same recognition as the guy that had beautiful code: does it solve the real world problem or not?

Safety critical, secure, etc, sure, but most is not. And, even with those, the libraries/APIs/etc are separate components.

Yup, because LLM inference can be scaled by adding racks of hardware. Consulting can't be.

Generative AI has failed to replace SaaS so far...It has disrupted plenty of other lower verticals in writing, proof reading, translation, graphics design, tutorials, searching case law etc. Unless the progress stops, you can't assume LLM efficacy has hit a ceiling.

Agents, properly setup can partially accomplish what you described already.

Their competitor is Google, which has already baked AI into search and allows you to talk to Gemini separately as well. Google already makes a hundred billion a year from text search and won't stop a free Gemini which adds value to their search. What makes you say that?

Also Deepseek and Alibaba would love to capture OpenAI users.

Large software platforms with their own ecosystems will be spared this fate. Mostly because AI can only create small and trivial apps and further model improvement in context size or otherwise may not be forthcoming so easily.

So large software platforms, think Jira/Confluence, MS Teams, SAP etc are not affected. But AI will definitely eat the solo dev SaaS, especially those handling trivial use cases.