The AI Intent Engine: From Plain-English Goals to Verifiable Network Checks

The gap between "the fabric should be fully meshed in AS 65001" and the hundreds of CLI lines that implement it is where outages hide. The AI Intent Engine closes that gap: you state the goal, it compiles the checks, and it tells you precisely where reality diverges.

How intent becomes checks

You express a desired property — a peering relationship, an MTU value, an NTP source — and the engine turns it into a plan of concrete, device-scoped checks. Each check runs through the same execution pipeline as the rest of the platform, producing a finding whether it passes or fails.

• Declare — describe the intended behaviour in plain language or the intent DSL.
• Compile — the engine expands quantifiers and emits a previewable check plan.
• Execute — checks run against live devices and record structured evidence.
• Verify — a pass/fail report keyed by device shows expected vs. observed values.

Evidence over assertions

A failing check is not a dead end — it shows the actual observed value next to the expected one, so the fix is unambiguous. A passing run is just as valuable: it is a green baseline you can point at during a change window to prove the network still does what you said it should.

Continuous verification

Schedule an intent and the engine only raises a finding when the result changes — continuous assurance without a flood of "still passing" noise. That makes intent the quiet backstop that catches drift the moment it appears.