AI for low-connectivity operations: design principles

A lot of AI is designed in rooms with fiber-optic internet, then shipped to places where the network is a rumor. In Indonesian operations — ships at sea, remote sites, plants on the edge of coverage — the gap between those two worlds is where AI projects go to die.

The fix isn't "better connectivity." It's designing for the reality: the field is offline-first, and the AI must work there anyway.

Principle 1: Capture must work offline

If data capture requires a connection, the crew will stop capturing. Period. Every form, photo, and record must be designed to work fully offline on the device — with sync that happens automatically when connectivity returns. The design rule: the field experience never depends on the network.

Principle 2: Sync is a first-class feature, not an afterthought

Offline capture creates a sync problem: conflicts, ordering, duplicates. A robust design needs clear rules for what wins when two devices update the same record, and a queue that survives bad networks — retrying in the background without the user ever touching it.

Principle 3: The agent works on what it has

Don't design an agent that needs the cloud to think. Keep the intelligence where the work happens: local models for classification and extraction, cloud only for heavy lifting and coordination. The agent should degrade gracefully — doing its job with whatever data and connectivity exist at that moment.

Principle 4: Audit-ready, even at the edge

Low connectivity doesn't mean low accountability. Every AI decision — even one made on a ship with no signal — must be logged, reviewable, and attributable to a human who approved it. When the connection returns, the full audit trail syncs.

We build for that reality. Prototype-first, offline-first, humans in the loop — on infrastructure that stays up even when the network doesn't.