Work /WorkBrain

Zoro.com — Case Study

WORKBRAIN

Keeping up with meetings and project context shouldn't be a second full-time job. So I built a system that does it for me. It now sits underneath an org-wide plugin.

Role

Lead Experience Designer. Gen AI Champion.

Timeline

2025 to 2026

Company

Zoro.com, a Grainger company

30 MIN

Two complete service blueprints for loyalty

~90%

Context filled before the first working session

57→2

Open questions cut to two for the steering committee

6

Active projects tracked at once with one sentence of input

Context shouldn't be a second job

When I built WorkBrain I was managing six projects at once: the HITL tool, where I was coding the front end, a design system reboot that needed a lot of decisions and discussion early on, a loyalty program that had just kicked off, feature refinements for my product team, a new promotion type that needed its own infrastructure, and a side project nobody had asked for, turning my notes into documentation other people could use.

Every one of those projects has meetings, and every meeting gets transcribed, but nobody wants to reread a full transcript to find the handful of comments that matter. Especially not me.

I can take notes in a meeting or I can be present in the meeting, and while I know a few people who can do both, I'm not one of them. Given the choice, I'd rather speak up, ask questions, and get the answers I need. That works fine on one project with a few meetings, but on six, the things I heard on Tuesday were gone by Thursday, and a decision we made three weeks ago was something I was pretty sure we'd made, even if I couldn't tell you when or where.

The context existed. It was just spread across transcripts, Slack threads, Jira tickets, Confluence pages and my memory, and keeping all of it straight had become a job in itself.

I already had one at home

I'd already solved this problem once, for restaurants.

At home I run a second brain named Gary, which lives on an old Mac Mini in my basement. I tell it something from my phone, and it files it, fills in the details, and gives it back to me months later when I ask. I built it so I'd stop forgetting book recommendations and actually remember them when I'm standing in the aisle at the bookstore. So the question was obvious: if Gary can remember where I wanted to eat, why am I the one remembering what the steering committee decided?

One folder, one sentence

WorkBrain is a folder on my laptop, with one folder per project, one shared to-do list, and one folder that matters more than the rest: the inbox. A rules file at the top tells Claude what to do with anything new, what it's allowed to rewrite, and what it can only add to.

Everything goes in the inbox, from meeting transcripts to decks, docs and exported threads. I don't name them well, I don't sort them, and I don't add any context. I just tell Claude Code one sentence:

There are new files in the inbox.

From inbox to record

  1. Drop

    A meeting transcript lands in the inbox.

  2. Match

    Claude checks it against the list of active projects.

    No match It stops and asks me before filing anything.

  3. Update

    It opens that project's context file and updates it under two rules.

    Rewritten every time

    Summary, open questions and people, so they always say where things stand now.

    Added to, never edited

    Key decisions and history, as dated entries. The record of what changed and why.

  4. To-dos

    Action items go to the shared to-do list, tagged with the meeting they came from.

  5. File

    The transcript moves into that project's transcripts folder.

  6. The payoff

    Next time, Claude works from the context file, not the raw transcripts.

All of this runs from one sentence: there are new files in the inbox.

If a file doesn't match a project, it stops and asks me before filing anything. That's the whole interface, and I designed it that way on purpose, because a system I have to maintain is a system I'll stop using in two weeks.

What it keeps

Every project gets one document, a context file, and it follows two rules. The summary, the open questions and the list of people get rewritten every time, so they always say where things stand now. The key decisions and the history only ever get added to, as dated entries that are never edited or deleted. Action items from every project go into one shared to-do list, which sits on my desktop in Übersicht, so I see them without opening anything.

The folder

  • CLAUDE.md

    The rules: what to do with a new file, what can be rewritten, and what can only be added to.

  • projects.md

    Every active project, used to match a new note to the right folder.

  • todo.md

    One shared to-do list across every project, carried forward until it is checked off.

  • _inbox/

    Where new transcripts and notes land before they are filed.

  • projects/

    • loyalty-program/

      • context.md One file, two rules

        Summary, open questions and people are rewritten every time. Key decisions and history are only ever added to.

      • transcripts/

        Every processed meeting, filed here once it's in the context file.

  • archive/

    Finished projects, moved here intact.

Plain markdown files. No database, no app. Every project folder has the same shape.

That second rule is the one that earned its keep. When a new meeting contradicts something already on the record, Claude flags it right there, next to the decision it contradicts, and adds it to the open questions. On Loyalty, it caught two important conflicts.

The first was the deadline. One meeting said the work was due in late summer, another said the end of the year, and a third called the launch date a suggestion. Each one sounded fine in the room, but side by side they didn't agree.

The second was bigger. One executive asked me to design and test a simple flat reward, while another asked me to plan toward spend tiers and unlocked milestones, and those are two different programs. When someone questioned it, I had a dated record of both requests.

I never read these files myself. Claude reads them for me and pulls out what I need, and it has more detailed context than I do, which is the point.

Showing beats saying

Saying "there's a conflict" doesn't move anyone, but showing it does.

So I had Claude pull every conflict out of the context file and into a shared spreadsheet listing every conflicting request, who said it, and the date they said it. I took it to my product manager and proposed a path: get aligned between the two of us, bring it to the steering committee, and set the record straight. Because the sheet had names and dates, we knew exactly who to go back to once a decision landed.

That sheet became the project's open questions list, where each row has the question, what needs deciding, and a status of open or final. It started with 57 open questions, and over a few weeks we worked it down to 12. We're still using it.

We walked product and design leadership through those 12, one row at a time, which left five for the full steering committee. That same day, three of us built the deck, with the question, why it matters, and what we recommend for each one. In a room of 20 decision-makers, five questions became two.

One of those was settled by a prototype. The question was where customers redeem their rewards, and our CFO had proposed the account page. I thought that was a detour customers didn't need, so I built a working prototype that redeems at checkout as a line-item discount. He clicked through it and said, "I'm sold. Seeing it like this and being able to click through and use it, this makes way more sense than what I was proposing." When I showed the same prototype to legal, it met their requirement too.

Loyalty's open questions

  1. 57The first list
  2. 12After a few weeks of working it down
  3. 5After product and design leadership
  4. 2After the steering committee
Every row had a question, a name and a date, so every answer had someone to go back to.

It's just files

When I built WorkBrain, Zoro didn't have Claude yet, so the first version ran on Codex. A few weeks later Claude arrived and I moved it over, and for a while I ran it from Cursor too, because that's where my budget was.

It survived all three without a rebuild, because there's nothing to rebuild. WorkBrain is just folders and documents, and any agent can read them. I run it on Claude Code now, since one tool is simpler than three, but anyone else who uses it can choose the agent they prefer.

Context is the fuel

Once every project had a current, honest record, a second thing became possible: I could hand that record to other tools.

I built seven skills that read from WorkBrain and produce the artifacts every project needs: the project brief, the user journey, the service blueprint, the competitive analysis, the open questions list, the Lean UX canvas, and the information architecture. They all work from one source of context, and they also read the documents the other skills produce. I built each one to run in Claude Code and in Cursor.

None of those skills are impressive on their own, because asking Claude for a project brief is futile if you can't give it up-to-date details to work with. They're good because of what they're reading.

Loyalty put it to work

I lead the loyalty program, the biggest project I've been on at Zoro, and leaders here call it the biggest project Zoro has ever taken on. That means a lot of stakeholders, and a lot of direction that didn't always agree.

When it came time for a service blueprint, I opened a blank Figma page, handed over the template, and let the context do the work. Thirty minutes later I had two, one for MVP and one for the long-term vision, and each one marked what was known, what was assumed, and where the gaps were.

That got us roughly 90% of the way there before anyone sat down in a room. I color-coded the assumptions, so the working session wasn't discovery, just a short list of questions with the right people present.

In October, the direction of the project shifted during an in-person summit, and I had about a week with almost no working time. It was only doable because the notes, the research, and the decisions were already organized and ready to be repackaged.

30 MIN

Two complete service blueprints

~90%

Filled in before the first working session

57→2

Open questions cut to two for the steering committee

It outgrew my laptop

I built WorkBrain for myself, and then I showed it to one teammate. He took WorkBrain and my seven skills and rebuilt them to run in Google Drive and Confluence instead of locally, because that's how he works.

That tooling became the foundation of a plugin that walks a designer through an entire project, phase by phase. He added skills of his own for other tasks, and a researcher added a couple for research. The whole digital experience team has access, and five of us use it so far. I then refactored the plugin so it runs in Cursor and Codex as well as Claude.

I didn't set out to build infrastructure. I set out to stay organized across six projects.

Part of the conversation

I walk into meetings prepared now, with a recap of the last conversation and what's still open.

On my first day back from a week off, it worked through the seven meetings I'd missed, and I walked into the next one caught up.

Then I give the room my full attention, and I don't worry about writing the important thing down, because I know it's captured.

It's like having a full-time assistant who sits in every meeting and keeps track of everything that matters to me, so I get to be part of the conversation instead of an observer of it.