Picture a business owner giving everyone access to an AI tool.
One person starts using it constantly. Another tries a few prompts but gets answers that need too much editing. Someone else carries on as before because they cannot see where it fits into their day.
Everyone has access. They do not yet have a shared way of working.
“Use AI more” leaves the important questions unanswered. Use it for which task? With what information? What should the finished work look like? Where does it go afterwards?
I would rather start with one job people recognise than a broad instruction to experiment.
Start with work they already understand
Consider a coaching business sending follow-ups after client sessions.
The coach finishes a call, sorts through their notes, writes a recap and sends the agreed next steps. On a busy day, that follow-up slips behind the next appointment.
The useful starting point is not “build a coaching agent.” It is helping the coach finish that follow-up while the conversation is still fresh.
Give them a repeatable way to turn approved session notes into a draft. Show an example of a good recap. Make clear where the coach checks it and where the finished message is sent.
A colleague should be able to look at the instructions and understand the whole job:
Use these notes. Produce this kind of recap. Check the commitments. Send it through the usual channel.
That is more useful than a folder containing impressive prompts with no explanation of when to use them.
It also makes training concrete. Instead of demonstrating everything the tool can do, work through a normal session together. Let the coach try the process on their next call. Ask which part helped and which part still made them stop.
Perhaps the draft is too formal. Perhaps the notes are missing the agreed actions. Perhaps copying information into the tool takes longer than writing the message.
Those details tell you what to improve. A polished demonstration cannot tell you whether the process fits someone’s working day.
Make it usable without the person who built it
A useful workflow should not depend on one enthusiastic colleague being nearby.
Keep the instructions beside the task. Include a finished example, the information needed to begin, and a clear place for the output. If something changes, someone needs to maintain those instructions.
For the coaching follow-up, that might mean a short guide in the workspace the coaches already use. They should not have to remember which chat contained the prompt or ask the owner to find it again.
OpenAI’s AI value-model guide puts practical workforce fluency at the foundation of deeper AI adoption. My interpretation for a smaller business is straightforward: help people use AI confidently in familiar work before asking them to manage a more elaborate system.
That does not mean every task needs AI. If a saved template does the job, use the template. If the real delay is waiting for someone to make a decision, faster drafting may not address it.
Nor does everyone need to become a prompt expert. The aim is for the person doing the work to understand enough to use the process, recognise a poor result and explain what needs changing.
Once that way of working is useful and repeatable, you can decide whether connecting tools or adding an agent would remove further effort.
Start with the job. Make it easy to do well. Then build around what people actually use.
An AI login opens a tool. A usable workflow gives someone a reason to return to it.