Some of my earliest Sidekick work involved setting up AI for five business owners with very different jobs. One lived in meetings and follow-up. One had long documents to compare. One had years of useful material that rarely made it into marketing. Another was starting a new role. One was building a business on the side.
All five had already tried AI. What they needed was a way to connect it to work that already mattered. That early work still shapes how I decide where an AI setup should begin.
They did not need more AI
Each owner could already open ChatGPT or Claude and ask a useful question. The problem was what happened before and after that question.
The right information lived in documents, meeting notes, email, software, or the owner's head. A good answer might help once, then disappear inside a chat. The next task began from another blank box. The owner still had to remember when to use AI, find the sources, explain the situation, judge the answer, and move the work forward.
Access to AI was not the missing piece. The missing piece was a path from real work to a useful result.
That is why a list of clever prompts rarely fixes adoption by itself. It may improve a conversation. It does not decide where that conversation belongs in the work.
Five pressures, five starting points
I approached each setup with the same question: what keeps taking too much time, energy, or attention? The answers led to five different starting points.
| Pressure | What I configured | What the evidence supports |
|---|---|---|
| Meeting follow-through | A repeatable path from notes and background to draft follow-up and actions | The workflow was built. No lasting time or quality result was measured. |
| Document comparison | One workspace for asking questions across selected reports | The setup existed. Repeated use and accuracy were not measured over time. |
| Content reuse | A repeatable drafting path built from existing material | The path was configured. It did not prove a marketing result. |
| New-role reference | A focused workspace for training notes, reference, and preparation | The system was delivered. Ongoing adoption was not measured. |
| Business exploration | A desktop project with relevant background and goals for planning | I observed early use, not long-term value. |
Once the work was clear, the setup was often simple: give the AI the right background, create a repeatable place to begin, define the useful output, and state what a person must check. Sometimes that meant a focused project inside a desktop app. Sometimes it meant a reusable instruction and a small set of source documents.
I now start most individual setups in familiar desktop products for the same reason. They are easy to open, relatively inexpensive, and simple to maintain. A custom system can do more, but more capability is useful only when the work earns the added cost and care.
What changed in my method
Training matters, but training alone does not connect AI to recurring work. Someone still has to design the path. More advanced does not always mean more useful. If a desktop workspace can prepare the work and leave the decision with the owner, that is enough to test the idea in real use.
Context is helpful, but it is not magic. Old information can be wrong. A clear instruction can still produce a weak answer. Important work still needs a person to check the facts and decide what happens next.
Most importantly, building something does not prove that it works. A polished workspace can sit untouched. A rough setup can be useful if it solves a recurring problem and is easy to reach at the right moment. Real work is the test.
- Start with one repeated pressure point. One annoying piece of work gives everyone a clearer test than a broad personal AI system.
- Define the human boundary first. Decide who checks the facts, approves the message, owns the relationship, or makes the commitment.
- Prove the path before adding connections. Add another tool only when real use shows that the extra layer is worth it.
- Measure the work, not the novelty. Look for less rework, faster preparation, better follow-through, or a task you can now complete. A good demo is not a business result.
Where I would start if you already use AI
Do not begin by asking what else the model can do. Pick one part of your week and answer four questions:
- What keeps getting delayed, repeated, or handed back to me?
- What information already exists before this work begins?
- What would a useful first draft or prepared decision look like?
- What must a person still check, decide, or approve?
Build the smallest path from those inputs to that prepared result. Use it on real work. Keep it if it helps. Change it if it creates friction. Stop if the maintenance costs more than the problem.
That is the lesson I kept from those five early setups. People did not need a more impressive AI tool. They needed a practical bridge between the tool they already had and work they were already responsible for.
You can see how that principle grew into the system I now use to run Sidekick in What AI Handles Inside My One-Person Business.
