I haven’t yet seen where use of AI hasn’t made things worse.

• Agents: Every instance of AI for “customer service” I’ve run into basically just closed cases for invalid reasons, spouted bullshit, and was used to block interacting with a real human, trapping users in infinite loops (thanks eBay) or disconnecting on them (thanks PayPal)

• Security: M365 seems to be relying heavily on this now, which means that the same phishing email which was successfully blocked last week will randomly get through a few times this week for no apparent reason

*Coding: sometimes useful, but also points to do stuff like reference functions that don’t actually exist in the API/language being used