System‑Prompts, Domänen‑Guardrails und PII

What does it take to build AI that cannot simply go offline when the internet has a bad day? How do you bring large language models, agents and deepfake detection into a 40 year old networking company without breaking trust, security and compliance? And why might the biggest career risk today be ignoring AI rather than embracing it?

You can also watch this episode as a video podcast on my YouTube channel.

In this episode of AI to the DNA, the podcast for everyone who wants to really understand and apply AI, I sit down with Keith Griffin. Keith is a Cisco Fellow, one of the highest technical roles at Cisco, and at the same time VP for Collaboration, responsible for products like Webex. He has been shaping Cisco's AI strategy since 2016, co authored the company wide responsible AI framework and works across engineering, M&A and research to decide where and how Cisco bets on AI.

We talk about what really changes when you move from cool AI demos to mission critical enterprise software. Keith explains how Cisco thinks about outages, data residency and failover, why a lot of speech plus LLM experiments fail in practice, and how you design collaboration systems so that AI features run as close to the problem as possible, from GPUs in meeting room devices to AI ready data centers on premises. We look at deepfake risks in video meetings, live deepfake detection, brand and safety guardrails and why hope is not a strategy when you wire LLMs and agents into real products.

From there we go into the workplace itself. Keith shares why speech is coming back as a serious interface for work, how AI note taking and agentic follow ups can fix very human weaknesses like bad follow through, and what an "AI native" office might look like when devices, network and agents all work together. We discuss AI governance as a team sport between product, IT, legal and procurement, and why the key soft skill for almost every role will be asking better questions and iterating with AI.

I'm Christoph Magnussen and at AI to the DNA, I'm not just asking how to use AI. I’m asking why it works the way it does so you can gain insights on the nature of this technology. This podcast is for everyone who wants to move beyond tool-tourism and master the real craft of AI: Learning to think, work, and build side by side with intelligent systems!

• (00:00) - Hope Is Not a Strategy: Why AI Will Accelerate Work
• (01:02) - Introducing Keith Griffin: Cisco Fellow & AI Strategy Leader
• (04:55) - Miscalculations, Timing, and 18 Years of Tech Change
• (06:48) - Enterprise vs. Consumer AI: Outages, Resilience, and Trust
• (10:25) - Cisco’s First AI Breakthrough: From Noise Detection to Responsible AI
• (12:36) - Deepfakes, Guardrails, and Why Security Becomes an Arms Race
• (17:48) - Building Agents for Real: Trust, UX, and Why Single-Agent Comes First
• (21:58) - Speech, Hardware, and Edge AI: Why Not Everything Belongs in the Cloud
• (28:19) - AI-Native Workplaces: Automation, Augmentation, and New Work Patterns
• (33:11) - Productivity in Practice: From Weeks to Minutes
• (35:58) - The Biggest Mistakes Enterprises Make When Adopting AI
• (38:22) - Scaling AI Governance: How Organizations Keep Pace
• (40:18) - Explaining AI to Kids: Schools, Skills, and Future Careers
• (43:58) - System Prompts, Agents, and New Attack Surfaces
• (47:23) - Where Next? Why the Most Important Question Keeps Changing