From AI User to Workflow Designer
Updated: Aug 16
Every month, an onboarding and offboarding survey file landed with my team.
Approximately 800 rows. Structured responses. Open-text comments. Patterns across functions and geographies that could shape recruiting, learning, compensation, and the broader employee experience.
In 2025, we were collecting it.
In 2026, I wanted us to action it.
The problem was not that we lacked information. The problem was that no one had a realistic way to do the same analysis, at that volume, every single month.
So I built a Microsoft Copilot agent I call Exit IQ.
I designed it to segment the data by function and geography. Separate what was actionable for Talent Acquisition, Learning and Development, and Compensation and Benefits. Surface patterns. Flag emerging issues. Compare one month with the next. Produce an executive summary, a scorecard, and the top themes for human review.
Now I upload the file and press enter.
I do not write a new prompt. I do not explain the survey again. I do not rebuild the analysis from a blank chat.
The method is already built into the agent.
That is the shift from using AI to designing a workflow.
A prompt solves a moment. A workflow preserves the method.
There is nothing wrong with using AI to summarize an email, polish a paragraph, or clean up a spreadsheet.
Those quick wins are useful. They are also usually dependent on the person remembering what to ask every time.
If I open a new chat each month, explain the survey, paste another set of instructions, and manually reconstruct the final analysis, AI is helping with a task. The workflow still lives in my head.
A designed AI workflow moves more of that method into the system:
What starts the work
Which inputs are required
Which steps should repeat
What AI should do
Where human judgment is required
What the final output should include
What must be checked before anyone acts
The goal is not to remove the person from the process.
The goal is to stop making the person rebuild the process.
The question that changed how I look at work
Once I built Exit IQ, I started asking a different question:
What work is sitting on my plate simply because no one has redesigned it?
Most people do not need a list of 100 AI use cases.
They need to look at work they already own and notice what keeps repeating.
Look for something that:
Happens every day, week, or month
Starts with a recognizable input
Follows a pattern you can explain
Produces the same type of output
Takes enough time or effort to be worth improving
Still has a human who knows how to judge the result
Can be tested with approved tools and appropriate data
Survey analysis fit because the file kept arriving, the analysis had a recurring purpose, the expected outputs were clear, and the decisions still belonged to people.
Design the work before you build the agent
The agent was not the starting point. The work was.
Before I could build Exit IQ, I had to make the existing process visible.
Workflow element | Exit IQ example |
Trigger | The monthly survey file becomes available |
Input | Approved onboarding and offboarding survey data |
AI work | Segment, compare, summarize, rank themes, and format the analysis |
Human work | Validate conclusions, add context, decide what matters, and determine what happens next |
Output | Executive summary, month-over-month scorecard, and top themes for partner teams |
Quality check | Confirm the analysis is grounded in the file and does not overstate or invent conclusions |
That map matters more than the tool.
If you cannot explain the trigger, input, steps, output, and review point, you are not ready to automate the workflow. You are asking AI to operate inside a mystery.
The human work did not disappear
Exit IQ does the work that is repetitive and difficult to sustain manually.
It can scan the file, apply the same categories, compare groups, and create a consistent first analysis.
I still have to decide:
Whether the data is appropriate to use
Whether a pattern is meaningful or misleading
What organizational context the agent cannot see
Which conclusions need a subject-matter expert
Which partner should receive an insight
Whether the evidence is strong enough to support action
I am not outsourcing conclusions about people to a model.
I am changing where I spend my time.
Instead of using my attention to scan hundreds of rows, I can use it to challenge the patterns, connect the insight to the business, and decide what deserves action.
AI can do the work.
I decide which work is worth doing and whether the result is good enough to use.
Build the smallest version you can test
A first workflow does not need every feature you can imagine.
It needs one real input, one useful output, and one clear human review point.
This is the thinking behind my 1-1-1 Agent Method:
One conversation. One workflow. One agent built in under 60 minutes.
The hour is not a promise that the agent will be finished.
It is a constraint that keeps the first build small enough to test.
Start with a representative file or example. Give the agent the context, rules, and output you expect. Then run it.
Ask:
Did it follow the process you intended?
Did it miss context a person who knows the work would notice?
Did it make an unsupported claim?
Was the output useful to the next person in the workflow?
Did the human review happen at the right point?
What should stop the workflow and require another person?
A prototype becomes useful when you test the workflow, not just whether the technology responds.
Do not measure the workflow by speed alone
A faster process is not automatically a better process.
If the output creates more review work, hides important context, or produces conclusions no one trusts, you have only accelerated the wrong thing.
For Exit IQ, the longer-term question is not simply whether the file gets analyzed faster.
It is whether:
The analysis is more consistent from month to month
Important themes are less likely to be missed
Partner teams receive information they can use
Human reviewers can trace the conclusions back to the data
The organization makes better decisions because the information is no longer sitting untouched
I have evidence that the workflow makes a previously unrealistic monthly analysis possible.
The next level of evidence is sustained use and whether the insights lead to better action.
The move from user to designer
An AI user asks, "What can this tool do?"
A workflow designer asks, "Which work should change?"
An AI user looks for a prompt.
A workflow designer identifies the trigger, input, steps, decision points, and standard.
An AI user evaluates one answer.
A workflow designer tests whether the process produces useful work consistently.
That ability will outlast any one tool.
The interfaces will change. The model names will change. The buttons will definitely move five minutes after everyone learns where they are.
The ability to see work as a system will still matter.
Start with the work already on your plate
Choose one recurring task you know well.
Do not begin by asking which agent you should build.
Ask:
What keeps coming back, and why am I rebuilding the method every time?
Map the work. Decide what AI can own. Protect the human decisions. Build the smallest useful version. Test it against the real standard.
Every month, the survey file still arrives.
What changed is that the work no longer begins from zero.
The tools are everywhere.
The advantage belongs to the people who know which tasks to hand off and which to own.
Find the workflow worth redesigning. JenInAI helps teams identify high-value AI opportunities, define human and AI ownership, and build practical prototypes.
Comments