The Class That Taught Everything and Nothing

The Class

I recently attended an online workshop aimed at helping small business owners leverage AI. The topic matters, because small businesses adopting AI tools is exactly where clear, practical guidance is needed most. I went in with good faith and genuine curiosity about how the material would be presented to a non-technical audience.

In roughly ninety minutes, the session managed to touch on MCP agents, model parameters, HuggingFace, AI-generated websites, financial document analysis, email automation, and job displacement. There was no connective thread between any of it. Each of those topics represents weeks or months of dedicated learning, and the class treated them as a highlight reel. Surface-level enough to sound impressive, shallow enough to be dangerous.

In a previous session of the same series, attendees had apparently used AI to generate a retail website from scratch. Capability demonstrated. Responsibility skipped.

What Was Said

1. Everything was covered. Nothing was taught. The session jumped between topics that each require their own dedicated curriculum. Model parameters, open-source model repositories, agent frameworks, website builders, financial tools, all in one sitting, with no depth on any of them. The audience left having heard of twelve things and understanding none of them. That’s not education. That’s a demo reel.

2. Live credentials were on screen. Access tokens and environment files were visible during the demonstration. There was no discussion of what those credentials do, how they should be stored, or what happens when they’re exposed. For an audience of small business owners learning by watching, this is the workflow they’ll replicate. Tokens pasted in plain text, shared in screenshots, committed to repositories they don’t understand.

3. “If you’re having trouble with your financial statements, you can upload them and ask AI.” That was the guidance. No caveat about where that data goes once uploaded. No mention of whether the platform retains it, trains on it, or who else can access it. No distinction between a consumer chatbot and an enterprise tool with data handling agreements. Just upload it and ask. To an audience of people who will now paste their bank statements into the first chat interface they find.

4. Thirty minutes of filler. The session included an extended tangent on job displacement, vague economic commentary, and general anxiety about the future of work. None of it was educational. None of it was actionable. It filled time that could have been spent teaching one thing well.

What Wasn’t Wrong

Here’s the part that makes this worth writing about: nothing the instructor said was technically false. AI can help with financial analysis. You can build websites with it. HuggingFace is a real platform with real utility. MCP agents are a thing. Model parameters do matter.

The problem was never accuracy. It was responsibility. Every capability was presented without its corresponding risk. Every tool was shown without its safety context. The gap between “AI can do this” and “here’s how to do this without exposing yourself” was never acknowledged, let alone addressed.

That gap is where the damage happens.

What Should Have Been Said

Data Sanitization

If you’re going to use AI to help with financial analysis, and it genuinely can help, you don’t hand it your raw bank statement. You strip identifying information first. Account numbers, routing numbers, names, addresses. All of it comes out before anything goes into a prompt.

Give the model the structure and the numbers it needs, not the document itself.

What not to do:

“Here’s my bank statement from March. Can you help me figure out where my money is going?”

That prompt, paired with an uploaded PDF, hands over your account number, routing number, transaction history, merchant names, and personal details. All of it is now sitting in a system you didn’t read the terms of service for.

What to do instead:

“I have the following monthly expense categories and totals: Rent $2,400, Supplies $680, Software $320, Contractor payments $1,500, Meals $210. What patterns should I look at to reduce overhead for next quarter?”

Same question. Same usefulness. No sensitive data left the building. The difference between those two approaches is the difference between using AI responsibly and handing your financial identity to a system you don’t control.

Prompt for Method, Not Output

Instead of handing AI your data and asking for answers, ask it to teach you the approach.

What not to do:

“Here are my Q1 financials. What should I change to be more profitable?”

You’ve uploaded real numbers into a third-party system, and you got back a one-time answer you can’t reuse without doing it again.

What to do instead:

“What’s a standard approach to categorizing small business expenses for quarterly review?”

Or:

“What are the most common overhead reduction strategies for a service-based small business with 3-5 employees?”

Or even:

“Walk me through how to build a basic monthly cash flow tracker in a spreadsheet.”

None of those require your actual data. All of them give you a framework you own and can apply yourself, repeatedly, with your real numbers staying local. This shifts the AI from a data processor to a tutor. You learn something. Your data stays private. The output is a methodology you can reuse, not a one-time answer that required uploading sensitive documents to get.

Credential Handling

If you’re demonstrating tools that use API keys or access tokens, even in a teaching context, you have a responsibility to explain what those keys represent. They are authentication credentials. They grant access to services, data, and billing accounts. Showing them on screen teaches the audience that displaying them is normal. It isn’t.

Environment variables exist for a reason. Keys should be stored securely, rotated regularly, and never visible in a shared screen, a screenshot, or a repository. When a key leaks, the consequences range from unauthorized API usage and unexpected charges to full data exposure. The audience will replicate what they see demonstrated. Demonstrate it correctly.

Scope and Depth

If you have ninety minutes with a non-technical audience, teach one thing well. A focused class on “using AI to organize your small business expenses,” with live examples, safety guidance, and hands-on practice, is infinitely more valuable than a tour of the entire AI ecosystem.

Depth builds competence. Breadth builds false confidence. An attendee who leaves knowing how to safely use one AI tool for one specific task is better prepared than someone who’s heard of twelve tools and understands none of them.

The Actual Problem

The instructor in this class wasn’t an outlier. They’re the norm.

Of the typical users I encounter, in workshops, in consulting, in casual conversation, most float through AI adoption with zero consideration for data privacy. The reasons vary. Some is genuine ignorance; they don’t know what happens to data once it’s submitted. Some is indifference; the tool is convenient, and the risk feels abstract. Some is the perennial “I don’t have anything to hide” position, which misunderstands what’s at stake entirely.

The people teaching AI to non-technical audiences are, in many cases, operating at the same level of awareness as their students. They know enough to demonstrate the tools. They don’t know enough, or don’t think to, explain the risks. The blind leading the blind, and the tools are sharp.

There’s no corrective mechanism operating at scale here. The incentive structure rewards adoption speed, not safety. Nobody is rewarded for pausing a demo to explain credential management. Nobody is penalized for telling a room full of small business owners to upload their bank statements to a consumer chatbot.

The gap isn’t malice. It’s absence. Nobody in the room thought to ask “where does that data go?”, including the person at the front of it.

The Point

The tools work. The education around them doesn’t. Not yet.

The gap between capability and comprehension is where the damage happens, and it widens every time someone teaches “upload it and ask” without teaching “sanitize it first.” Every time credentials are shown on screen without context. Every time twelve topics are surveyed in ninety minutes and the audience leaves feeling informed when they’re actually more exposed than when they walked in.

This isn’t about gatekeeping AI from small business owners. They should be using these tools, the productivity gains are real and meaningful. It’s about giving them the same safety awareness that professionals take for granted. Sanitize your inputs. Protect your credentials. Ask for methods, not miracles. Learn one thing well before moving to the next.

The tools aren’t the problem. The gap is. And right now, the people positioned to close it are widening it instead.