In my company I have a lot of people think the same thing.. very senior folks thinking AI agents "are bad, they cannot build real code, etc", which is funny to me cause I've been building tools nonstop with it - mostly using amp and cc - and they use those tools which I've builded with AI 100% - no edits on my part -.. so I don't know..
sometimes they're doing stuff i haven't thought they can do, and failing on stuff that feels super obvious.
and it's not yet clear how does one grasp that frontier, like whether the one should try harder with prompting, with another agent -- or the issue is with something more fundamental, and it's 5x faster to go the old fashioned way for this particular issue
because the frontier keeps changing, and not necessarily improving in every spot!
I’m enjoying to read your thoughts a lot.
I want to be a good writer like you !
Excellent analysis! How to bridge that expert-user divide?
In my company I have a lot of people think the same thing.. very senior folks thinking AI agents "are bad, they cannot build real code, etc", which is funny to me cause I've been building tools nonstop with it - mostly using amp and cc - and they use those tools which I've builded with AI 100% - no edits on my part -.. so I don't know..
agent frontier is very, very jagged.
sometimes they're doing stuff i haven't thought they can do, and failing on stuff that feels super obvious.
and it's not yet clear how does one grasp that frontier, like whether the one should try harder with prompting, with another agent -- or the issue is with something more fundamental, and it's 5x faster to go the old fashioned way for this particular issue
because the frontier keeps changing, and not necessarily improving in every spot!
I enjoyed reading that issue.
And your pub speech was probably so spot on that it hurt.