I recently joined Cisco TechBeat for a conversation with Aaron “AB” Belinfanti and Pascal Bornet about AI, agents, leadership, learning, and the ways organizations are trying to keep up with a technology landscape that seems to change every day.
The conversation covered a lot of ground, but I kept coming back to one point: leaders can’t guide AI adoption from the sidelines.
That does not mean every executive needs to become a machine learning engineer. But leaders who are making decisions about AI strategy, investment, governance, risk, and organizational change need enough direct experience with the tools to understand their texture.
The same is true for practitioners. Reading about models, tools, and agents is useful, but real intuition comes from trying to build something with them: seeing where they help, where they fail, and where the demo stops matching reality.
In other words, understanding AI requires direct experience.
The Bike-Riding Problem
I used an analogy in the conversation that gets at this pretty well: you can’t learn to ride a bike by reading a book.
You can read about balance, steering, braking, momentum, and body position. You can watch videos. You can study someone else’s technique. But eventually, you have to get on the bike, wobble, overcorrect, scrape your knees, and develop the feel for it.
AI is increasingly the same way.
You can read blog posts and model cards. You can follow the benchmarks. You can listen to podcasts and watch demos. Those things are useful, but they are not a substitute for building something, pushing the technology into your own context, and developing judgment about what works, what breaks, and what is just theater.
This is especially true with agents. The interesting questions are rarely answered by a demo. They emerge when the system has to interact with real tools, real data, real users, real permissions, real edge cases, and real organizational constraints.
That is where the learning happens.
Leading From the Console
For leaders, hands-on exposure is not about becoming a power user for its own sake. It is about developing better judgment.
A leader who has spent time with Claude, ChatGPT, Gemini, or another capable model—using it for real work, iterating with it, hitting its limits, and seeing where it creates leverage—will have a different conversation with their team than one who has only read about the technology.
They will ask better questions. They will have a better feel for what is plausible. They will be less likely to confuse a polished demo with a production-ready system. And they will be better equipped to create the conditions under which their teams can experiment productively.
The Opportunity Is Bigger Than Chatbots
Another distinction I tried to emphasize is that there are at least two important ways to think about using AI.
The first is interacting with AI directly: asking questions, brainstorming, summarizing, researching, drafting, coding, and using AI as a kind of reasoning or productivity partner.
That matters, and it is often the easiest place to start.
But the second is, in many ways, more powerful: using AI to create tools, workflows, and systems that change how work gets done.
This has been one of the biggest shifts in my own work. AI is useful when I interact with it directly, but the bigger impact has come from using it to help build systems around my workflow.
At TWIML, that has meant automating large parts of our podcast production process, creating internal tools, stitching together services, and turning repeated work into reusable workflows. In many cases, the final tool is not “AI” in the visible sense. It does not necessarily look like a chatbot. But AI helped make it possible, and that distinction matters.
For many organizations, the first wave of AI adoption has been about giving employees access to tools. The next wave will be about helping teams rethink the workflows those tools make possible.
That is where the leverage is.
Intent, Iteration, and Taste Still Matter
Of course, using AI more does not automatically mean using it well.
We are already surrounded by AI-generated slop: generic text, shallow analysis, synthetic-looking visuals, and content that carries the statistical shape of insight without the substance.
The answer is not to avoid the tools. It is to use them with more intention.
AI works best as part of a process. That process still needs direction, judgment, iteration, and taste. The user still has to know what they are trying to accomplish, what good looks like, what context matters, and when the output is drifting away from the point.
The mistake is treating AI as a one-shot content machine. I call this the AI slot machine fallacy: put in a prompt, pull the lever, and hope something useful comes out.
The opportunity is treating it as a collaborator in a larger creative, analytical, or operational process.
That means bringing your own ideas to the table. It means asking the system to challenge your assumptions, generate options, clarify tradeoffs, or help turn rough intent into something more structured. It means iterating. It means rejecting what is bland or wrong. It means staying in the loop.
The more powerful these systems become, the more important that loop becomes.
TWIML’s Own Evolution
In the conversation, AB asked about the origin of the TWIML AI Podcast.
The short version is that I started it nearly ten years ago as a forcing function. I was trying to keep up with a field that already felt like it was moving quickly. I would end the week with dozens of open browser tabs—papers, articles, demos, announcements—and the podcast was a way to make myself process what I was seeing and share what seemed important.
In retrospect, those were quiet times.
Today, the challenge is not finding enough to talk about. It is making sense of a field where the volume of change can easily exceed our ability to absorb it.
That has pushed me to think more deliberately about TWIML’s role. The show has always been about conversations with people doing important work in AI: researchers, engineers, founders, enterprise leaders, and practitioners. But increasingly, I see the work as helping people connect those conversations to the questions that matter now:
What is real?
What is changing?
What is hype?
What should builders and leaders pay attention to?
What does it take to put these systems to work responsibly and effectively?
That mission feels more relevant than ever.
Key Takeaways
A few ideas from the conversation are worth underscoring:
- Leaders need direct exposure to the tools. They do not need to become engineers, but they do need enough experience to understand the technology’s capabilities, limits, and organizational implications.
- AI is moving too quickly for passive learning. Reading, watching, and listening are useful, but real intuition comes from hands-on use.
- Agents raise the bar for practical understanding. Their value and risk only become clear when they interact with real workflows, tools, data, and constraints.
- The biggest opportunity is not just using AI as a chatbot. It is using AI to help create new tools, workflows, and operating models.
- Good AI work requires direction, judgment, iteration, and taste. The AI slot machine fallacy is expecting useful output from a one-shot prompt instead of treating AI as part of a larger process shaped by human intent.
Thanks to AB and the Cisco TechBeat team for having Pascal and I on the show!
You can watch the full episode here: