2 things:
AI isn't smart
Yet current AI is enough to transform employment as we know it
We often hear that AI isn't real intelligence or that it doesn't think for real. Not hard to verify if you interact with popular LLMs. They forget context, make things up, etc. So, clearly, no superintelligence.
That said, the ability to simulate intelligence from best AI models (Claude 4, GPT o3, Grok 3, DeepSeek R1) is already good enough to outperform humans on some cognitive tasks.
Now, what makes people good at a job is not ONE task, but an ability to perform hundreds of varied tasks, switch context, remember context, having soft skills, etc.
So I think that claims like "software engineers are over" etc. are not true, they lack nuance on what makes a good professional. Coding agents don't know what they do and can't finish their job.
Human beings still have an edge today because they can finish a work they started, remember context, and adapt to situations.
However, this edge has reduced and I think mid-term (3-5 years) it's likely to reduce more. Why?
Because you don't need superintelligence or to fix hallucinations for AI to replace knowledge workers. It just needs to simulate intelligence very well (so faking understanding, collecting more context data and retrieving it when needed) and cost significantly less than labour.
Which will give many employers the choice between AI tooling vs hiring someone/contractor. Where hiring was required it will become optional.
Will AI software replace the worker? Not exactly, but enough for the worker not to be absolutely needed anymore. Which means less ability to negotiate (salary, home office, etc.).
Rather than replacing everyone, AI will eat the competitivity of knowledge workers, and we won't live in a golden age anymore (salaries always increasing, more geographical freedom).
People very good at their job (good software engineers doing proper engineering, project managers, designers, writers, compliance) will still have opportunities but large-scale, average people or entry-level will have less opportunities, I believe.
Because AI will be good enough to simulate average performance (i.e. doing intern job). Not to replace a superstar head of engineering.
This is a system issue and not an individual one. So you can try to increase edge as much as you want by learning to work with AI or AI tooling, I think short-term it protects, but long-term it's about economic incentives from corporations.
I think States are not taking this questions seriously, and that maintaining competitivity of knowledge workers is key. In countries like France, this would mean reducing charges an employer has to pay to hire someone (not the salary, let's be clear). Otherwise economic incentives for replacement will be too big.
Conclusion: even not real intelligent AI means reduced competitivity of knowledge workers (except people really good) which means less golden age of employment (jobs, raises, freedom).