Mike's Head Is Your Wiki.

Mike has been on the same roof for two hours. His phone has rung four times. All four were the same question, asked by three different dispatchers about three different jobs.

Mike is the senior tech. Mike has been there fourteen years. Mike knows which model of Carrier needs the breaker pulled before the panel comes off, which neighborhoods still run on the old copper, which customer’s dog will bite even though the wife says it won’t. Mike is, functionally, the company’s knowledge base.

He’s also on a roof.

The dispatchers are good. They’ve been there long enough to know they should ask Mike before they tell a customer something. They’re not the problem. The owner isn’t the problem either. The owner has been promising himself for six years that “we should really document all this.” He even started a Google Doc once. It died at four bullet points.

Here’s the thing. The wiki isn’t missing. The wiki is Mike. And the bug is structural, not lazy.

The wiki you’ve already built without meaning to

Every shop I walk into has the same shape. There’s a Mike. There’s a Sarah at the front desk who knows which crew can be paired on a hot day and which two will end up swinging at each other by lunch. There’s an office manager who memorized the workaround for the scheduling software the owner refuses to replace. There’s a sticky note on the inside of a binder that says “NEVER promise Tuesday before noon to Mrs. Helman, see April incident.”

That whole shop, every shop like it, is running on tribal knowledge. The technical name is institutional knowledge. The honest name is if Sarah quits we lose six months. On Reddit, every dispatcher horror story ends the same way. The dispatcher leaves. The shop discovers, in a panic, how much of the business lived in one person’s head and three custom spreadsheets nobody else opened.

Yeah. That’s the bug.

The bug isn’t that nobody wrote it down. The bug is that writing it down has always been the most expensive thing in the room. The owner who’s “been meaning to document our intake process for years” isn’t lying. He’s pricing it correctly. Documenting takes hours per process, the documentation goes stale within a month, and the person who’d benefit from the doc is the new hire who isn’t there yet. So the doc never gets written. Mike stays the wiki. The roof keeps the wiki.

That math has held for thirty years. The math just changed.

What a developer figured out that an operator hasn’t

Last month a developer ran a small experiment. He created a folder. He dumped raw materials into it. Papers he was reading. Conversations he was having. Half-finished thoughts. He pointed an AI at the folder and told it to build a wiki. Just build it. Read what’s there, write articles that organize it, link the articles to each other, and update the wiki every time something new gets dropped in.

In a few weeks the wiki had ~100 articles and 400,000 words. Indexed. Cross-linked. Self-maintaining.

That’s a developer story. It is also, almost word-for-word, the operator story you’ve been failing to execute for a decade. Folder of raw materials is your email. Your Slack. Your text threads with Mike. The 2019 Google Doc nobody can find. The voicemails. The scribbled note that says “NEVER Mrs. Helman before noon.”

The raw materials already exist. They’ve existed the whole time. What was missing wasn’t the content. What was missing was someone to organize it, maintain it, update it, and answer questions from it without that someone being an actual person on actual payroll.

That someone now exists. It costs $20 a month. The excuse is gone.

The Professional Recipe has an Ingredient that solves exactly this

We talk about AI deployment around here using a framework called The Professional Recipe. Seven ingredients. Every AI station in your business needs all seven to hold up. Training. Context. Guardrails. Examples. Output Over Process. Measurement. Feedback Loop.

Ingredient #2 is Context. What’s true about this customer, this job, this moment. Most operators skip Context. Not because they don’t believe in it. Because building it felt expensive.

That’s the part that just changed.

Context used to mean writing down everything an AI station might ever need to know about your business. Procedures. Customer history. Pricing logic. Why Mrs. Helman gets afternoon-only. Why Mike pulls the breaker first on the older Carriers. Doing that by hand looked like a six-month project, so it became a six-year project, so it became a decade-long project nobody ever started. The shop kept running on tribal knowledge because the alternative was a documentation marathon nobody had time for.

Now you don’t write the wiki. You seed it.

You take the existing raw materials, the ones that have been sitting in your business for ten years, and you point AI at them. The AI reads your email threads with Mike. It reads the dispatch logs. It reads the workaround Sarah documented in a Slack DM in March. It reads the customer notes in your CRM. It reads the four bullet points the owner wrote in the dead Google Doc. It writes the wiki.

Then every new question that gets asked, every new email Mike sends from a roof at 2 p.m., every new “never promise Tuesday morning to Mrs. Helman” moment, gets dumped into the same folder. The wiki updates itself.

Context, the ingredient most shops have shipped with completely blank, just got cheap. Literally cheap. The thing you’ve been pricing at six months of one person’s time is now a Sunday afternoon and twenty bucks a month.

Why this isn’t a chatbot

I want to head off the pattern-match before it happens.

This isn’t “set up a chatbot.” It isn’t RAG, which is the developer term for re-discover the answer every time you ask. It isn’t a help desk. It’s not a tool you use; it’s a thing your business has. The way a body has muscle memory.

The right way to think about it. Mike’s head is the wiki right now. The wiki is read-only, located in one human, and that human is on a roof. Goal isn’t replace Mike. Mike is great. Mike has fourteen years of judgment that AI will never have. Goal is pull the wiki out of Mike’s head and into a place every dispatcher can ask without putting Mike on speaker.

Mike still gets to be Mike. Mike just stops being the bottleneck for which neighborhood is still on copper.

Right? That’s the move.

The dispatcher asks the wiki. The wiki answers. If the wiki doesn’t know, the wiki asks Mike, captures Mike’s answer, and now the wiki knows. Permanently. Mike doesn’t have to remember to write it down. Mike answers a question once. The wiki ate it. Six new dispatchers can ask the same question for the next ten years and Mike never hears it again.

Does that make sense? The wiki gets smarter every time it’s used. That’s the part that’s new. That’s the part the developer figured out and the operator hasn’t.

The Monday Move

Don’t try to build the whole wiki this week. That’s the same trap the dead Google Doc fell into.

Pick the one question your team asks most often that always routes to the same person. The one where the dispatcher closes their laptop, walks across the office, and says “hey Mike, real quick…” Pick that one.

Write Mike’s answer down. Five sentences. Then add the five most common follow-ups your team always asks after the first answer. Five more answers. Five sentences each.

You just started your wiki. Drop the file into a Claude project, a custom GPT, a NotebookLM notebook, whatever your shop already pays for. Tell the AI: “This is our shop’s knowledge base. When my team asks you something, answer from this. When you don’t know, ask me, and I’ll add it.”

Then for the next thirty days, every time someone asks Mike a question, paste Mike’s answer into the same folder. Every email Mike sends a dispatcher with a fix, paste it in. Every voicemail Sarah leaves about Mrs. Helman, transcribe it, paste it in.

In thirty days the wiki answers thirty questions before anyone has to ask. In a year, it answers most of them. Mike still gets the calls that need Mike. He stops getting the calls that need the wiki Mike happens to be storing.

That’s the gap that just closed. That’s Ingredient #2 going from blank to filled.

So.

The knowledge exists. You just never wrote it down. Now you don’t have to.

Mike is the wiki. Mike is on a roof. The dispatcher who needs the answer at 2 p.m. on a Tuesday doesn’t need Mike, she needs what Mike knows. For thirty years there was no way to separate those two things, so we hired more Mikes and prayed nobody quit.

There’s a way to separate them now. It costs less than the smallest tool in your stack. It runs on raw materials you’ve been generating for a decade and never organized. The first version takes a Sunday afternoon. The compounding version takes a thirty-day habit.

Pick the one question. Write the answer. Add five follow-ups. Point AI at the folder.

Watch the wiki start answering before anyone has to ask.

Source influence: a developer pattern of pointing an LLM at a folder of raw materials and letting it build and maintain a persistent, interlinked knowledge base. Distilled and operator-translated.

Framework spine: The Professional Recipe, Ingredient #2: Context. Read the full framework