A while back, I helped out as a small hand at a team building for a profession where everything they handle is client-confidential — every file is something that can't leave the room.
I'd assumed an industry this protective of secrecy would keep its distance from AI, worried about data leaking out. The opposite is true — they're quietly bringing it into daily operations.
The phrase they kept repeating
What stuck with me was one line they kept coming back to: "human in the loop."
AI doesn't run the job — it takes the repetitive tasks, and a person still looks, still decides. AI drafts and organises; the one who signs off and approves is always human.
The real question: where does it run?
Watching from the side, one thing kept turning in my head.
For a field where everything rests on data confidentiality, the real question isn't whether to use AI — it's where it runs.
Public AI (OpenAI, Gemini, Anthropic and the like) means the data has to leave the building. But for them, whether data can leave the building at all is the red line.
So I keep wondering if the next step for industries like this is local, on-prem deployment — AI running on their own machines, data never stepping outside.
I don't have the answer. Genuinely asking — if you're in a data-sensitive field (accounting, legal, medical, finance): is on-prem AI inevitable, or am I overthinking it?