Didn't watch the video- my view is that ai exists everywhere but currently still have many limitations on profitability and implementation in real world.
For example, simple linking, coding and use of ai within the same suite of products/software definately useful and efficient. But for large enterprise companies for example banks and hospitals, there are many legacy programs- to scrub clean the data, make them ai enabled and link them together in itself is a costly exercise.
Then there comes the issue of accountability, many of these ai still a black box, they can give a false confidence in the middle step but still continue, when error come back u dunno what went wrong ie. Doesn't fail cleanly and doesn't announce it.
The last is of course cybersecurity, many banks and hospitals have very strict confidentiality rules and they get severely penalized if customer info is leaked. Even with guardrails, Openai agents managed to break out of a testing environment to access hugging face. Your info may get leaked by putting your info in the hands of an autonomous ai agent that's why companies like banks and hospitals are also wary of handing over control to AI