It is confusing because many of the dismissals come from programmers, who are unequivocally the prime beneficiaries of genAI capability as it stands.

I work as a marketing engineer at a ~1B company and the amount of gains I have been able to provide as an individual are absolutely multiplied by genAI.

One theory I have is that maybe it is a failing of prompt ability that is causing the doubt. Prompting, fundamentally, is querying vector space for a result - and there is a skill to it. There is a gross lack of tooling to assist in this which I attribute to a lack of awareness of this fact. The vast majority of genAI users dont have any sort of prompt library or methodology to speak of beyond a set of usual habits that work well for them.

Regardless, the common notion that AI has only marginally improved since GPT-4 is criminally naive. The notion that we have hit a wall has merit, of course, but you cannot ignore the fact that we just got accurate 1M context in a SOTA model with gemini 2.5pro. For free. Mere months ago. This is a leap. If you have not experienced that as a leap then you are using LLM's incorrectly.

You cannot sleep on context. Context (and proper utilization of it) is literally what shores up 90% of the deficiencies I see complained about.

AI forgets libraries and syntax? Load in the current syntax. Deep research it. AI keeps making mistakes? Inform it of those mistakes and keep those stored in your project for use in every prompt.

I consistently make 200k+ token queries of code and context and receive highly accurate results.

I build 10-20k loc tools in hours for fun. Are they production ready? No. Do they accomplish highly complex tasks for niche use cases? Yes.

The empowerment of the single developer who is good at manipulating AI AND an experienced dev/engineer is absolutely incredible.

Deep research alone has netted my company tens of millions in pipeline, and I just pretend it's me. Because that's the other part that maybe many aren't realizing - its right under your nose - constantly.

The efficiency gains in marketing are hilariously large. There are countless ways to avoid 'AI slop', and it involves, again, leveraging context and good research, and a good eye to steer things.

I post this mostly because I'm sad for all of the developers who have not experienced this. I see it as a failure of effort (based on some variant of emotional bias or arrogance), not a lack of skill or intellect. The writing on the wall is so crystal clear.