Most AI advice is a sales pitch. You need the opposite — a straight read on what actually works in your environment, what it really costs, and what to ignore. AI removes the natural brakes on a business, which makes operating discipline matter more, not less. This is not a practice that runs on hype.
Three things are true about AI at work in 2026 — and the people selling it rarely say them out loud.
The right tool is decided by the environment you already run — Microsoft, Google, or a regulated boundary — not by which model is “best.” The answer changes with your setup.
The compliant, mission-safe tiers cost more and often run older models than the consumer tools. The more sensitive the data, the further behind the frontier you sit.
Business-tier usage caps quietly throttle the everyday work you’d actually automate — the spreadsheets, the PDFs, the inbox. “Give everyone a subscription” hits a wall fast.
What is the work — and how sensitive is the data? Answer those two, and “it depends” becomes a plan.
1 · What is the work? Three kinds of AI, not one.
Drafting, analysis, brainstorming. Low risk, no hands on your systems.
Does the office task — email, documents, spreadsheets, slides. Touches your operations.
Writes and refactors software. The most capable — and the widest blast radius.
The further right you go, the more of your data flows through the tool — which pushes the work up the sensitivity tiers below.
2 · How sensitive is the data? Four tiers, and each one changes the answer.
The specific, current tool-by-tool answer for your environment — which vendor, which boundary, what it costs, what’s authorized this month — changes constantly. Keeping that map current, and matching it to your work, is what I bring to an engagement.
AI in a business isn’t “give everyone ChatGPT.” It’s an operating model: which work runs on which tool, at which sensitivity level, in your environment, within your budget — with governance for the data that needs it. I map your actual work onto the tiers, recommend what to buy and stand up at each level, and tell you plainly what to skip. It stands on its own, or folds into a fractional operating engagement as the AI-governance layer.
I run governed, private AI myself — my own practice’s analysis and content pipeline runs on a self-hosted AI stack on my own hardware, so client work never leaves my boundary — and I’ve hit every limit I’ll warn you about. The goal isn’t adoption for its own sake — it’s what actually helps, matched to the job, proven in practice, and governed for the data that matters. AI accelerates the work. It doesn’t replace the judgment, or the discipline to check that “done” is actually done.
A short fit check will tell you where AI genuinely helps you — in your environment, within your budget — and where it doesn’t.
Start with a fit check