Ep. 213: What AI Should Never Do, Enterprise Scaling, Governing AI & Navigating IT Roadblocks

A Q&A episode grounded in real business pain

Paul Roetzer and Kathy McPhillips frame this as the 18th edition of their AI Answers series, pulling unanswered questions from SmarterX's free Intro to AI and Scaling AI classes, which have now served more than 60,000 people. They also plug two SmarterX initiatives: a May 14 walkthrough of their 2026 State of AI for Business report based on 2,000-plus survey responses, and a new AI for Business Boot Camp kicking off July 16 in Columbus for 180 attendees.

How cautious companies get unstuck from “policy mode”

The first theme is the tension leaders feel when they know AI matters but keep treating it like an IT problem. Roetzer's advice is practical: don’t wait for perfect governance, find responsible experiments that avoid sensitive data, and build the case with something small and obvious — like using Writer, Jasper, or ChatGPT for marketing outputs that only rely on public information.

What actually creates momentum: vision, peers, and one clear use case

Asked whether tools, strategy, or leadership pressure matter most, he says the tools are already good enough; what’s missing is understanding what they can do and a CEO-level vision that says, “We are going to figure this out.” For skeptical employees, he thinks peer-to-peer influence works best: seeing a nontechnical coworker use AI successfully is often more persuasive than any formal training program.

The security nightmare around agents is already here

One of the sharpest moments comes when Roetzer talks about people casually giving AI agents logins because “it kept getting hung up,” including access to Chrome and other systems. He says security teams should be alarmed, but also argues companies can’t simply slow down, because AI-native startups are moving too fast — which creates a brutal tension for legacy companies, especially in healthcare and banking.

Stop trying to learn everything; get excellent at one platform

For individuals, especially early-career professionals, his advice is to focus. Pick one assistant — Claude, Copilot, Gemini, or ChatGPT — learn its core capabilities deeply, understand advanced models and deep research, and use that to transform your actual work instead of spiraling over every frontier experiment you haven’t tried yet.

Build now if smarter models make your product better, not obsolete

On the “should I wait?” question, he borrows Sam Altman’s test: if a better model will obsolete what you’re building, don’t build it; if a better model will enhance it, go now. That applies whether you’re building an internal tool, a startup, or a Claude project, and he contrasts that with brittle automation stacks that are likely to get wiped out by rapidly improving general-purpose models.

HR, hiring, and the growing AI-vs-AI dynamic

Roetzer says HR is full of practical use cases — recruiting, onboarding, note summarization, assessments, performance reviews — but also warns the human element is getting harder to preserve. Candidates are increasingly using AI to create polished resumes, cover letters, and interview prep, so hiring teams need to assume there’s an “AI facade” and work harder to understand the real person behind it.

What AI should never fully take over — and why younger generations may be different

His most personal answer is about writing: he believes AI is already a better writer than most people, but still refuses to use it for his editorial newsletter, LinkedIn posts, podcast opinions, or keynote writing because that’s the part of the work that feels human and meaningful. He closes with a story about his 13-year-old building a school project using Minecraft scenes, iMovie, and AI-generated crowd noises, and you can hear both the amazement and the thesis: if kids are taught to use AI responsibly instead of as a crutch, their ability to learn and create could race ahead.