Two Ways to Talk to a Smart Tool

Post 3 of 5 — There are two kinds of AI. We are teaching our team both.

So far we’ve talked about how our work is changing and how we’re adapting—now let’s talk about how we actually use AI.

Most people I talk to think of AI as one thing. It is not. There are two distinct flavors of AI my team is learning to work with, and the difference matters for the quality of the work you receive from us. I want to take a few minutes to explain both, because the distinction is genuinely useful, and most other professional advisors in your world have not articulated it yet.

The first kind: generative AI

This is the AI most people have heard of. ChatGPT, Claude, Gemini. You give it context and direction, and it generates something like a draft analysis, a memo, a structured comparison, a research summary.

The catch is that the quality of what it produces depends almost entirely on the quality of how you set up the conversation. Garbage in, confidently wrong garbage out. So we have trained the team on five principles for working with generative AI well:

• Establish role and context. We do not ask AI to “analyze this.” We ask it to analyze “as a tax advisor reviewing a SaaS company’s Q3 financials, with these specific concerns.” Role definition activates the right domain knowledge.
• Provide explicit quality criteria. We tell the AI what to prioritize and where to be cautious. “If you are less than 75 percent confident in any conclusion, say so explicitly.” This is how we keep AI from giving us confidently wrong answers.
• Request structured outputs. “Give me the analysis as a three-column table: Finding, Evidence, Recommendation.” Structured outputs are easier to validate and integrate into your work product.
• Use examples to calibrate. We show the AI what good looks like by giving it examples from past work. That trains it faster than any number of paragraphs of instructions.
• Require transparency about uncertainty. We ask the AI to flag every assumption it is making and rate its confidence. This turns it from a black box into a transparent collaborator.

These five principles are part of what my team practices with Chad, the custom GPT I introduced in the last post. They are also why two firms using the same off-the-shelf AI tool can produce wildly different work. The tool is the same. The skill is not.

The second kind: semantic AI embedded in your software

The second flavor is newer and, in some ways, more interesting for our daily work together. It is the AI that lives inside the software you and we already use. These tools do not generate. They understand. You ask them a natural-language question about your actual financial data, and they give you the answer because they have direct, secure access to it. No keyword search. No clicking through six menus to find the right report. No exporting to Excel and pivoting.

You can just ask: “What did we pay this vendor over the last twelve months, and is that pattern unusual compared to similar vendors?” And get an answer. And these questions are becoming more securely embedded in our financial tools online that we all use.

I sometimes describe the difference this way. Searching software the old way is like looking for one specific ingredient on a grocery store shelf. While talking to embedded semantic AI is like having a chef who knows the whole pantry walk you through what is possible to make a creative dish.

Why both matter to you

My team is getting fluent in both. Generative AI is how we draft, structure, and stress-test our analysis before it reaches you. Semantic AI inside your software is how we, and increasingly you, query your own financial reality faster and more conversationally than was possible eighteen months ago.

In some engagements, we are starting to set you up to use the embedded tools directly. You can ask a question about your own payment history without waiting for us to pull a report. That is not us trying to step out of the loop; it is us giving you instant access to information you should have always had instant access to. The judgment work; what the answer means, what to do about it, how it fits the larger picture, is still where we earn our price.

The firms that get the most out of AI are the ones whose teams know which kind of AI they are talking to and why. Ours do. That is part of what your price is buying right now.

If you want to talk about what any of this means for your engagement with us, message us at [email protected] to set up a time to chat.