The Work Stays Inside

Perplexity’s Portable Computer is easy to misread.

It runs AI locally. It asks before using the cloud. It starts on expensive Nvidia hardware most teams will not buy soon.

So the lazy reaction is: interesting demo, not relevant yet.

Wrong lesson.

The useful part is the boundary. Some work can leave the building. Some work can leave only after the sensitive pieces are removed. Some work should not leave at all.

That decision cannot live in a chatbot policy nobody reads.

Agents make the question sharper because they do more than wait for pasted text. They read files, compare records, draft answers, pull history, and move across systems. Once that happens, privacy stops being a data-only issue. It becomes a question about the shape of the work.

A customer escalation is not the same as a market scan. A board draft is not the same as a blog outline. Finance notes, contract terms, pricing exceptions, internal strategy, and client-specific context need different rules.

The first move is boring and useful: map ten recurring AI workflows. Put each one in a lane.

Cloud is fine. Cloud after redaction. Private/local only. Not for AI yet.

That map will matter more than the hardware.

The companies that get this right will not be the ones with the fanciest local machine on a desk. They will be the ones where people know, before the agent starts, whether the work is allowed to leave the room.