Stop Starting Over With AI: Turn One Good Conversation Into a Better Way of Working

At the Hawai‘i Women in Tech “How I AI: Members Edition” panel, we talked about some impressive ways people are using AI. We also talked about the experiments that flop, the risks we need to question, and the very human judgment technology cannot replace.

As I reflected on the conversation afterward, one thought kept coming back to me:

The best way to get started with AI is simply to have a conversation. But that is only the beginning.

The bigger opportunity is to learn from that conversation, capture what worked, and make the next attempt better—without starting from scratch every time.

Start with one moment of friction

People often ask, “What should I use AI for?”

That question may be too big.

Instead, look closely at your day. Where did you lose time? What did you postpone? What task do you repeat? Where did you think, “There has to be an easier way to do this”?

Get micro.

Maybe you need to draft an email you have been avoiding. Turn scattered notes into a checklist. Review a requirement for ambiguity. Prepare questions for a stakeholder conversation. Even something as simple as I was pulling some of the transcripts from one AI app and pasting in a Word document to use with another AI app (that screamed process improvement!!). Or even when you’re online and just trying to find the two useful minutes buried inside a 20-minute instructional video.

Start with one real need—not a tool searching for a problem.

Have a conversation, not a prompting performance

You do not need the world’s most sophisticated prompt to begin.

Tell AI:

• Here is what I am doing.
• Here is what I am trying to accomplish.
• Here is where I am stuck.
• Here is what a useful result would look like.
• Ask me questions if you need more context.
• (And my favorite!!) I don’t know how to write the prompt for this, so I’ll need your help in writing the prompt first so that we can then complete this task.

And don’t just think staring at your computer. I actually use voice conversations during the 15- to 20-minute walk home after taking my daughter to school. I may talk through an article idea, shape an email, or organize thoughts I do not want to lose. But I start with a statement of need and then ask it if it’s okay if I just work through my thoughts (hopefully more than verbal vomit…but that’s why I like non-judgmental AI!).

The value does not come from writing a perfect prompt. It comes from adding the context that helps the technology support the work I am already doing.

Give feedback while you work

Now that you’re excited on the work you can do with AI, here’s where many of us now know that you have to give feedback. This is the step people will skip in their rush to complete their assignment, flustered that AI didn’t do it perfectly.

AI gives you something mediocre, so you fix it yourself and move on. When you do this, the tool never gets the benefit of your judgment.

Tell it what missed the mark:

• “That sounds too formal for me.”
• “You focused on the requested feature, but we have not clarified the business need.”
• “This example is closer to what good looks like.”
• “You missed two constraints that matter every time I do this.”
• “Before giving me another answer, ask me the questions you should have asked the first time.”

Your feedback is part of the work. It reveals what you know, what you value, and where human judgment matters.

And we often know to always verify the result. (TRUST…but VERIFY!) AI can help us think, draft, analyze, and move faster. It does not relieve us of responsibility for what we use.

The easiest way?  Whatever you take from AI, go back when you’re done with that task and give AI your finalized result. You asked for help with an email.  Go back and copy and paste your actual email sent so it can see the difference. Asked for slides for a presentation.  Go back and upload the final PowerPoint file so it can see what you did (andlearning point this week – double check ALL animations….not everything works in real-world as it does in the AI tool..SMH…). This one step alone can help train your tools to see what you expect. It many not be able to read your mind, but it can learn from what you do!  (Sounds like working with our stakeholders…can’t read their minds…but we can work much better with a backlog of user stories or process map….gotta get those ideas out and shared!).

Capture the learning before you leave

Now the power of lessons learned is to LEARN. Once you get a useful result, do not close the conversation and lose everything you just taught it.

Ask:

Based on our work together, what did you learn about how I perform this task? What instructions, examples, criteria, or reference materials should I save so we can produce a stronger result next time without repeating this entire conversation?

Then ask:

How should I set this up for repeated use? Should I create reusable instructions, a template, a project workspace, a checklist, or a specialized assistant?

Now you are doing more than completing one task. You are capturing lessons learned and improving the process.

You can also ask AI to separate:

• What it can do consistently
• What information it needs each time
• What should be standardized
• What still requires human judgment
• What should be checked before the output is used

That is how experimentation becomes a dependable way of working. And it’s simple – just ask IT!

Share the win—especially if it feels small

During the panel, I described someone who had set up Claude with a project charter, requirements templates, and strong examples. When a new user story arrived, the tool could help create a first pass at the requirements.

My immediate question was: Why aren’t you showing everyone else how you did that?

The response was essentially, “It’s just something I did.”

Those “little” experiments are exactly what organizations need to see.

Leaders are trying to determine how AI should be governed, where it creates value, and which uses are worth supporting. Meanwhile, employees are already discovering practical applications—but often keeping those discoveries to themselves.

Close that gap by sharing:

• The problem you were solving
• What you gave the AI
• What it produced
• What you corrected
• What still required your judgment
• What time, energy, or rework it saved
• What someone else would need to repeat it safely

That is more useful than simply announcing, “We used AI.”

Make space for more important work

AI is not automatically valuable because it produces something faster. The real question is what that saved capacity allows us to do.

AI now handles tasks my assistant previously completed. Yet she bills more hours than before because she is doing more valuable work—the work that needs her experience, initiative, and human judgment.

That is the outcome I want from technology.

Not more output simply because we can produce it. More space for people to think, connect, analyze, create, and contribute where they bring the most value.

So, where should you start?

Pick one small, recurring point of friction. Talk it through with AI. Verify the result. Give feedback. Capture what was learned. Set it up so you do not have to teach the same lesson again—and then share what worked.

One conversation can solve today’s problem.

Capturing and sharing the learning can improve how the work gets done from this point forward.