Making AI (net)work: tips for a successful AI-integrated network with Jason Gintert

Most AI network deployments look great in a clean POC environment — so why do so many stall out once they hit real production traffic and real teams?

In this session, Jason Gintert draws on his work with the US Networking User Association and his consulting practice, Bits in Flight, to lay out six fundamentals for AI deployments that actually survive contact with the real world — plus a bonus seventh on the human side of adoption.

He covers:

• Why “data before the model” is the single biggest predictor of failure — unaligned hostnames, timezones, and source-of-truth data derail more AI projects than model choice ever does
• Making AI show its work: a three-stage trust progression from validate everything, to spot-check, to delegate the repeatable — while keeping determinism between the AI and the actual network
• Meeting engineers where they already work — chat ops, ticketing integrations, and CLI overlays — instead of adding to dashboard sprawl
• Why trust has to be earned through visible wins, and why every AI deployment needs one named, accountable owner
• The “10-second rule” for frictionless correction and tuning, so the system visibly improves instead of repeating the same mistakes
• Picking measurable KPIs from day one so the project can survive budget scrutiny
• The human factor: recruiting AI skeptics rather than dismissing them, and why messaging AI as “superpowers, not replacement” determines whether adoption sticks

He closes with a practical 90-day framework — lay the foundation, run read-only with humans in the loop, automate one task, measure before/after, then expand — as a starting structure teams can adapt to their own timeline.