AI that works for your people, not instead of them.

The core problem

Tools like Claude read and write so fluently that every instinct says you're dealing with a person. You're not. Give a colleague of ten years a vague assignment and they fill the gaps with intuition — they know your organization, your members, what's going on this week. The machine has none of that. What you don't specify, it fills with the average of everything comparable ever written, and the average of every paragraph ever written is boilerplate.

That's why the same tool produces filler at one desk and excellent work at the next. The difference is never talent and never magic. It's how much of the gap the person closed with specifics before the machine had to fill it with averages. Closing that gap is a learnable, non-technical skill. I can teach that. And then the fun really begins.

AI 101 – Getting Started

The first hour

A one-hour session with your whole staff — skeptics especially welcome, because the skeptics become your quality-control department. Everyone reads a short document, then watches three live demonstrations on your real work. In the third one I make Claude fail on purpose, on screen, so your team sees exactly where the tool breaks before anyone is asked to trust it.

A trial with gates, not a leap of faith

Six weeks. Real tasks, each with a timed baseline. A shared prompt library, an error log of every fabrication caught in review, and a go/no-go gate between stages: internal work first, outward-facing drafts only after the team has demonstrated the review discipline. For a small team the software runs under $500 end to end, month to month. At week six you keep the uses that demonstrably earned their place and drop the rest without ceremony — and if the trial fails, you've bought a documented map of what AI cannot do for you, which ends the argument as decisively as success would.

Rules people can actually follow

A data rule that fits on an index card: public material — use freely; routine internal material — permitted, on a plan that doesn't train on your data; anything given to you in confidence — never entered, full stop. A review standard: nothing leaves the building without a named person reviewing it with accountability.

AI 201 – Next Steps

Once the habits are in place and the trial has told you which uses are real, the ceiling gets a lot higher. Some directions a next-step engagement can take:

• Using Cowork effectively. Most people use AI to draft things. Tools like Claude Cowork will actually do the work: whole tasks, start to finish. Handing work over safely takes real review discipline, and it's worth learning.
• Crafting AI policies. After six weeks of logged cases, you can write a policy grounded in your own evidence instead of a template downloaded from the internet. It will fit on two pages, and your board can approve it in one reading.
• Working with AI as a team. A shared briefing document, a few voice samples, and a common prompt library upgrade everyone's work at once. The skill should live in the organization, not in one enthusiast.
• Goal loops. Give an agent a task and a checklist, then let it grind against that checklist until the work is actually done. It will happily produce draft #53. It never gets irritated.
• Routines. Some work should simply run itself on a schedule: the weekly report waiting for you Monday morning, the follow-up sweep after every event. A person still reviews and signs everything. That part never changes.

And whatever your situation calls for. Unusual problems are the fun ones.

Subtle Risks

1. It invents, fluently. At the edge of its knowledge, an AI doesn't just make up facts — it can invent having checked them, in the same calm, credible tone it uses when it's right. There is no tone of voice that signals a fabricated phone number. So facts flow into the tool from you; they never flow out of it into the world unverified.
2. Machine-scented writing costs more than it saves. Unedited output has tells, and your customers detect them at increasing rates. A hollow letter tells a member they weren't worth a person's time — the exact inverse of what you're selling. Every draft gets a human edit, a detail only you know, and a real signature.
3. It should never decide. It has no stake, no accountability, and no knowledge of the people involved. It structures decisions — options, criteria, the case against — and a named person makes them. It needs review by an accountable human before e-mail goes out or a post goes up, and of course it can only advise in matters of law, tax, HR or safety.

The pocket card

The handout your team keeps after the first hour — the whole method, compressed.

Who this is for

Organizations whose value moves through written and structured work — startups, small and mid-sized businesses, legacy enterprises, nonprofits, professional offices, community organizations — where the binding constraint has always been hours, never skill or ideas. In a town Wausau's size, that's most of Main Street. I work in person across central Wisconsin and remotely beyond it, and I build everything to outlast me: the habits, the prompt library, and the policies belong to your staff, not to your consultant. Success means you no longer need me.

Get in touch

Wondering what AI is actually good for in your organization — including what it's bad at? That's a conversation worth an hour of your time. The first hour is free and the coffee or New Glarus is on me. Tell me what your organization does, which work eats the most hours, and where your team stands on AI today: enthusiasts, skeptics, or (most likely) both. I typically reply within one business day.

felixgallo@humanetic.ai