The Work Stays Inside
Local AI is a management test: decide which work can touch the cloud, which work needs approval, and which 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.
The useful part to examine now 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.
Customer escalations, market scans, board drafts, and blog outlines carry different risks. 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 know, before the agent starts, whether the work is allowed to leave the room.