By "useful" he means "will make VC's and uninformed investors feel good"... gotta remember, the definition of "useful" is relative.
Realistic examples of small, useful AI projects a junior dev could do in a sprint:
1) check for spelling/grammar mistakes on company website 2) summarise transcripts of zoom calls 3) scan codebase for misleading comments 4) generate mermaid diagrams from terraform 5) Scan Jira for near-duplicates on a backlog and add a label
None of these require data science teams to get involved.
that’s a lot of cash saved. basically as good as cash earned, when you’re getting funding rounds of cash.