Researchers are finding new ways to get these AIs to do more things, to generalize their application, but I do not think the actual usable, general intelligence they can use to do those things is growing substantially.
They are amazing tools, but AIs need more-general intelligence to fully automate software development, and that will require different AIs, not more Terawatts.
Who will then define, debate and hold standards/regulations/norms, then? and adherence to those standards, as well as the legal responsibility that comes with it?
What happens when a totally automated software is the main cause of the destruction of private property, or the death of even one single person?
Who is going to answer for these? (that is, pay, fix and possibly go to jail)
How are those legal consequences going to pressure the automated development industries to comply to safety/control rules?
If models could take over all meaningful work there would be nothing left for humans to do but war.