You hired an ace on Monday. Smart, fast, willing. What do they need to actually be useful? You’d tell them what the business does and who your customers are. You’d forward them a few emails you’d sent recently, so they could draft in your voice. They’d see your current projects, your client list, the patterns you actually cared about. None of that would feel optional. It’s just how onboarding works.
You should think about AI the same way. Most owners I work with don’t. They open ChatGPT, Claude or Copilot, ask a question, take the answer and close the tab. Some of them tell me “I tried ChatGPT but gave up.” Some say “I know I should be using AI but I don’t know where to start.” Most say a version of “I’m paying for it and using it for one thing a week.”
Meanwhile the tools have changed underneath them. ChatGPT and Claude now remember you between conversations. Both connect to your email, files and calendar. Both can take a whole job, like “pull last month’s invoices and build me a summary sheet,” work through it step by step and hand back a finished file. A year ago that meant a separate product you had to go and find. Now it ships in the app most owners already pay for.
Most owners don’t have an AI problem. They have an onboarding problem.
The new features make onboarding matter more, not less. A new hire with access to your inbox needs better instructions than one who only answers questions. What follows is the playbook I take every Solo client through before we build anything custom. I first wrote it in May 2026 and updated it in October, because two of the habits I used to teach by hand are now built into the tools. Most of it you can apply this week.
Get the input right
The cheapest move is still the biggest one. Stop describing your situation. Show it the actual thing.
Most people type three lines describing an email thread, get a generic template back, and conclude AI is shallow. Give it the actual email thread, the actual document, the actual customer complaint, and the output goes from generic to something you’d actually send. You wouldn’t tell a new hire “write a response to a difficult client” without showing them the email. Same rule.
Often you don’t need to paste anything now. Connect your email and Google Drive or SharePoint once, and “read the last three emails from this client and draft a reply” just works. On a business plan, your IT provider usually approves those connections first. Pasting is still the quickest route for a one-off.
Then beat the question into shape. “What are some marketing ideas for my business” produces sludge. “I run a six-truck residential HVAC company, my biggest pain is that 30 percent of my maintenance plan members aren’t renewing in year two, give me three retention email scripts that don’t feel like upsell pressure” produces something you can run on Monday. Specificity isn’t a stylistic choice. It’s the difference between an answer and a script.
The practitioner-grade move I’d add: tell it what you’ve already tried. Most people walk into the conversation as if it’s the first time anyone has ever thought about the problem. Two minutes spent on “here’s what I’ve already tested, here’s why each one didn’t land” cuts the back-and-forth a lot. Almost nobody does this, and it’s the cheapest skill increase available.
Teach it once, then check what it learned
The old advice here was to write yourself a handbook, because the chat box forgot you every time. That part has changed. Memory now comes switched on for most personal ChatGPT and Claude accounts. On team and business plans an admin decides whether it’s available, and you may still need to turn it on yourself.
That ends the Monday-morning reintroduction. It also creates a quieter problem. The AI now builds its picture of you from whatever you happen to mention, and it won’t tell you when that picture is wrong. Mention a price you’ve since changed, or a client you’ve since lost, and it may keep working from it.
So onboard it on purpose. Spend one conversation telling it what you’d tell a new hire: what the business does, your role, your customers, how you write, and your defaults. “When I ask for emails, keep them short and direct. When I ask for analysis, be skeptical.” Then ask it to remember all of that.
Then check what it learned. Once a month, open the memory settings and read what it thinks it knows about you. Both let you review what they’ve stored and delete what’s wrong, and Claude lets you edit each memory directly. Fix what’s stale. It takes five minutes, and it’s the difference between an assistant that knows your business and one confidently working from last spring.
Projects still earn their place. Memory is one general picture of you. A project in Claude or ChatGPT keeps one client’s or one job’s files and rules together, so they don’t leak into everything else.
Voice is the other side of context. “Write in my voice, casual but professional” almost never works. Five examples of your actual writing and “match this register” almost always works. You wouldn’t tell a new hire “sound like me.” You’d forward them five emails you’d actually sent.
Then save the routines that worked. When an output hits, don’t just save the prompt. Turn it into something you can run by name: a skill in Claude, a custom GPT in ChatGPT. Six months of this and you have a real toolkit you run by name.
One more habit: correct the misses out loud. Most people only keep the wins. When an output misreads the situation, say what was wrong and ask it to remember the correction. “Don’t open client emails with a compliment. Remember that.” With memory on, a correction you make once usually sticks, so the misses are worth more than they used to be.
Use it as a thinking tool, not an output machine
Most people ask AI for an answer. The real move is asking for a critique of your answer.
“What am I missing?” “What would the opposite case look like?” “Steelman the argument I’m not making.” Most casual users have never pointed AI at their own decisions. They use it as a vending machine. Insert prompt, retrieve snack, judge AI on whether the snack was good. You hired a genius. Genius is wasted on snacks.
Same problem on drafting. Most people ask AI to write the first draft and end up rewriting it anyway because the voice is wrong and the structure is generic. Flip the order. Write a rough first draft yourself and ask: “What’s weak here? Where would a skeptical reader push back?” You keep control of voice and substance. The AI runs the review pass.
And iterate. The single line “tighter” produces dramatic results. So does “more specific.” So does “now write it for a skeptical reader” or “cut the hedging.” Most casual users take the first output, judge AI on it, and stop. Two or three passes is where it starts to surprise you. The first output is the warm-up.
The practitioner-grade move I’d add: don’t ask for opinions, ask for trade-offs. Generic “what do you think?” gets fortune-cookie answers. “What does this trade away?” gets operator-grade analysis. That shift turns an advice machine into a thinking partner.
The boring 80%
Most people fail at AI by starting with the strategic 20 percent. The investor letter. The board memo. The high-stakes copy where every word is second-guessed. They get a draft that doesn’t quite land, decide AI “isn’t quite there,” and stop using it.
That’s like firing a new hire for not nailing their first board memo. The real wins come from the recurring drudgery first: meeting notes, inbox triage, data cleanup, standard email replies. The work where the bar is functional, not exquisite, and where the volume eats your hours every week. Build trust on low stakes. Move up from there.
The companion move: layer AI into the way you already work. If you already write meeting notes, then summarize them, then send a follow-up, keep that routine and have AI run the summarize step. People who redesign their whole process around AI tend to stall. The fastest adoption is the lowest-friction one.
Once a recurring job works by hand a few times, put it on a schedule. ChatGPT and Claude can both run a saved task every morning or every Friday: summarize yesterday’s customer emails, draft the weekly client update, flag invoices past 30 days. You set it up once, then read the result instead of producing it.
The habit underneath all of this: treat workflows as the unit, not prompts. Most casual users polish prompts. Power users improve workflows. A great prompt that runs once a month is worth less than a decent one that runs reliably twice a week. The thing you’re improving isn’t the sentence you typed. It’s the loop the sentence sits inside.
You already have the workspace
When I first wrote this, the last move was outgrowing the chat box: leaving a chat window for a separate AI workspace with memory, reusable skills and connections to your real tools. You don’t have to go anywhere for that now.
ChatGPT Work, now included on every plan in the ChatGPT desktop app, takes an outcome instead of a question. It pulls context from your connected tools, plans the steps and hands back a spreadsheet, a deck or a document. Claude does the same kind of work in its desktop app: give it a task and it works with your files and returns the finished result, and it uses the same memory as your chats. Asking and doing now live in the same app. Most owners just haven’t opened the second half.
A tool that can act needs what a new hire with system access needs: clear instructions, access to only what the job requires at first, and someone checking the work before it goes out. Start it on drafts it hands back to you, not on sending email or changing records. Widen what it can do as it earns your trust.
A year in
A year in, the work compounds in ways that are hard to see from the start. Three things separate a long-time power user from someone who just switched the features on.
Context becomes a library, not a memory. Memory holds small facts about you. A library holds the real documents: a file for each client, each project, each policy, each part of the business, kept current and marked when something is out of date. Most owners already have this scattered across Google Drive, SharePoint or Notion and never point the AI at it. Now that both tools can read those folders, connecting it takes a few minutes in settings. Keeping it tidy is what makes the AI fluent in your business instead of guessing.
Feedback loops become explicit, not vibes. A casual user judges AI output by reading it and going “that’s good” or “that’s bad.” A power user checks against a short written standard: a proposal gets scored before it goes out, a draft gets reviewed from the client’s point of view, a workflow that fails gets fixed and the fix gets written down. Over time the standards get sharper, and so does the AI’s work.
The job becomes the brief. Both tools can now break a big job into steps and work through them on their own. What separates a power user is what they hand over: what done looks like, what it may touch, and how the result will be checked. A vague brief split five ways gives you five vague results. A sharp one is how a small business runs at the throughput of a much larger one without adding headcount.
By a year in, casual users still type questions into a chat box. Power users hand it jobs.
Reading the playbook and shipping none of it is where most owners actually live. A blog post can’t do the setup for you.
Doing all of this yourself is the long road. Sidekick Solo is the shortcut. I set up Claude, ChatGPT or Copilot around your business within 48 hours, then coach you for 30 days. The clients who come to me usually start with “I just want someone to set it up for me.” If that sounds easier than building it yourself, the conversation starts with a call.
Free resource: Starter Prompts.
