The more I use AI, the more bullish I get. And I’m getting to a point where the value of what I understand about this stuff might be too valuable to share.

Not because I’m hoarding secrets. But because most people have shown zero interest in what AI can actually do. They stopped at “ask ChatGPT to update my docs, copy paste back into Word.” And for that majority, I genuinely don’t know what the next step looks like for them.

AI Skill Is Not SQL Skill

Here’s the thing people get wrong: they think learning AI is like learning SQL. Pick up a query language, add it to your resume, move on. It’s not that.

AI skill is closer to people skill. But hold your horses — it’s not the kind of “people skill” you put on a LinkedIn profile to sound like a leader. It’s a hybrid between your hard skills and your soft skills. It’s knowing what to ask, when to push back on the output, when to trust it, when to throw it away and start over. It’s judgment. And judgment doesn’t come from a tutorial.

The people who are good with AI are the people who were already good at thinking about problems. AI just made that thinking 10x more leveraged. If you couldn’t articulate what you wanted before, a chatbot isn’t going to save you.

The Corporate Data Problem

So let’s say vibe coding takes off. Replit, Figma, whatever — tools that let non-engineers build things with natural language. Cool. But how valuable is that when your corporate data sits behind a firewall? When your files are scattered across file servers, SharePoint, three different SaaS tools, and that one shared drive nobody admits to maintaining?

You can’t vibe code your way through a workflow when the AI can’t see half the inputs.

And then what do you do as a company? Hand over your data to OpenAI or Anthropic or some third-party AI vendor? That’s the question nobody wants to answer honestly. The enterprise AI pitch is always “we’ll handle your data securely” — but the moment you send your proprietary processes, your customer data, your internal logic to an external model, you’ve made a bet. Maybe it’s a good bet. But it’s still a bet.

UI Should Get Simpler, Not More Complicated

If AI does the bulk of the work, the interface should shrink — not expand.

Think about it. If an AI agent is doing the research, writing the draft, pulling the data, formatting the output — what does the human need to see? A read-only dashboard. Maybe. For people who care about watching the process. The AI doesn’t need a UI. It can’t see one. It works in text, in structured data, in function calls. Building elaborate interfaces for an AI agent is like putting a steering wheel on a train.

The real transition is from tables, dropdowns, and multi-step forms to a text box. That’s the new interface. And there’s a learning curve there that people underestimate. Going from “click through five screens to generate a report” to “describe what you want in a sentence” sounds simpler, but it requires a completely different kind of thinking.

For less technical users, human-in-the-loop UI still makes sense. But the design goal should be guidance, not complexity. Preset prompts. Guided prompt generation that helps users be specific without making them learn prompt engineering. Strict templates that save tokens and produce better results. The UI becomes a prompt coach, not a control panel.

Where This Is Going

The gap between people who use AI and people who work with AI is going to widen fast. The first group gets a slightly faster copy-paste workflow. The second group gets leverage that compounds.

Most companies haven’t figured out the data problem. Most tools are still built for the copy-paste crowd. And most people haven’t even started thinking about what their work looks like when AI can do 80% of it — because they’re still stuck on the 20% it can’t do cleanly.

The skill isn’t technical. It’s structural. It’s how you decompose problems, how you evaluate output, how you decide what to delegate and what to keep. And that’s exactly the kind of skill that doesn’t transfer through documentation.