Turning feedback into focus: using AI to fill gaps and amplify strengths
A follow-up to my earlier piece on “first impressions of Glean”, where I described experimenting with creating an agent to map feedback from peers, managers, and self-assessments to role profiles to help growth conversations be more structured and grounded.
The problem with feedback
Every review cycle, employees receive structured feedback in Workday, ratings, free-text comments, input from peers, managers, and cross-functional partners. It's genuinely rich data. But sitting with a spreadsheet of comments and trying to connect them to your role profile expectations is hard work. You're holding too many things in your head at once, and the synthesis you end up with is only as good as your energy levels on a given afternoon.
I wanted something that could do that synthesis more rigorously, not to replace reflection, but to make the starting point better.
What the agent actually does
The agent is called “Map Feedback to Role Profile”. A user exports their Workday feedback as a spreadsheet, uploads it alongside a link to their role profile and a note of their level. The agent reads the spreadsheet, extracts all free-text comments with their rater context, and maps strengths and growth areas to the specific competencies expected at that level.
It attributes feedback to the right people. It separates what's genuinely strong from what's a development opportunity. And it ends with a playful "superhero superpower" section, because not everything has to be serious, and because the strongest consistent theme in someone's feedback often tells you something true about how they show up.
After the report, the user can save it as it is or turn the full report into a manager-ready, promotion-supporting summary tailored for their manager, or turn it into a personal development plan that directly leverages the key feedback themes.
The feedback I have received from those I support, other managers and senior designers that used it to inform their goals have described this as helping feedback become more actionable and relatable to expectations, in addition to shifting AI into something closer to a coaching assistant.
Where it got interesting: trying to improve it
The agent worked. But the workflow had friction. To run it, users had to:
Export their feedback spreadsheet from Workday
Find and paste a link to their role profile document
Manually type in their level
Add their own self-assessment and manager feedback (optional)
Four inputs. For a self-serve tool you want people to actually use, is four opportunities for someone to give up.
So I started working through the automation question, not by writing code, but by using Claude to interrogate what was actually possible. I had hoped that Claude could automate far more intuitively than Glean but what I learned from Claude is that I could push Glean harder than I was. I discussed with Claude what data Glean already has, is Workday indexed and what the people directory contains.
What I found:
Workday feedback data isn't currently indexed in Glean, the spreadsheet export stays for now. Claude suggested that I ask users to copy-paste from the spreadsheet into the agent, rather than add the step to export it. I pushed back because the layout context would be lost and Clause agreed.
Claude suggested two other options that required IT support and integration which seemed unlikely at this stage (‘Connect Workday to Glean’, or ‘Connect Glean API + Workday API’).
However, the role profile documents are in Glean's search layer and the Glean people directory already contains every employee's job title and level.
Two of the three manual inputs can be eliminated. The agent can look up who you are, infer your level from your title, find your role profile document, and pre-populate both fields. The only thing a user needs to provide is the feedback spreadsheet itself.
I now have a prompt to add to the agent editor to make this happen. No engineers required.
Where it got interesting: trying to improve it
The agent worked. But the workflow had friction. To run it, users had to:
Export their feedback spreadsheet from Workday
Find and paste a link to their role profile document
Manually type in their level
Add their own self-assessment and manager feedback (optional)
Four inputs. For a self-serve tool you want people to actually use, that's four opportunities for someone to give up.
So I started working through the automation question, not by writing code, but by using Claude to interrogate what was actually possible.
The first suggestion was the obvious one: ask users to copy-paste their feedback directly into the agent rather than export a spreadsheet. Less friction, no file management. I pushed back. The Workday feedback display is a structured table with rater names, questions, ratings, free-text comments all in columns. Copy-pasting collapses that structure, and the agent relies on it to correctly attribute feedback to the right people. Claude agreed. First option off the table.
The next two options, connecting Workday to Glean, or integrating both via API, were technically sound but required IT involvement and infrastructure work that wasn't realistic in the short term.
At this point I could have stopped. The Workday export was the headline problem, and there was no quick fix for it. But I kept going, because the export wasn't the only friction. There were still two other manual inputs: the role profile link and the level confirmation.
Claude authenticated directly with Glean and ran a live search. What came back: role profile documents for every function are indexed in Glean's search layer. And the Glean people directory already contains every employee's job title, which maps directly to their level in the role profile framework.
The agent doesn't need to ask. It can look up who you are, infer your level from your title, find your role profile, and pre-populate both fields automatically. The whole interrogation took less than twenty minutes. The answer was already inside the system.
Two of the four manual inputs can be eliminated today, with a single prompt change. No engineers required.
The Workday export remains, for now. But the journey to that conclusion taught me something more useful than the fix itself.
What this taught me about AI fluency
The most useful thing about this process wasn't the output. It was the conversation that got me there.
I wasn't asking AI to be creative on my behalf. I was using it to map the edges of a system I didn't fully understand, asking questions, checking assumptions, discovering that data I thought was unavailable was already there. That back-and-forth is different from prompting. It's closer to investigation.
In my First Impressions piece I wrote that the value of Glean was helping me "reduce noise, improve clarity, and show up more intentionally for my team." What I'm finding is that AI fluency, for a design manager, is essentially the same thing applied to how you work with AI itself. The skill isn't knowing which tool to use. It's knowing what question to ask next.
Where this goes now
The Workday automation still has dependencies; there are data access and privacy questions I need to work through carefully before this becomes fully automated at scale.
But the agent is being used. People are getting value from it. And I understand the system it sits inside better than I did before.
Glean is still relatively new at Expedia. These examples will feel basic soon. I'm looking forward to that.
---
Senior Manager, User Experience Design at Expedia Group.