Human workflow data for AI
Real work.
The reasoning
behind it.
CommonWhy is building datasets of people doing their own computer work and explaining their decisions out loud. Hear what they noticed, what experience told them and what they still remember the next day.
On screen
Checks the shipping line in a spreadsheet.
Said out loud
This total looks off. I’ve done enough of these to know shipping often gets counted twice, so I’m checking that line first.
Our Windows recording prototype works. Member onboarding, uploads, review and delivery are being built. The first paid collections will be scoped with buyers.
01 / What we ask people to explain
How they work things out.
The screen shows what happened. Only the person can say why. Our Member Guide gives members specific things to say out loud as they work.
Read the full questions- Decisions
- Why this step, and why now? What were the options, and why this one over the others?
- Experience
- What tells them this is right? What have they seen before that makes them work this way, or spot that something is off?
- Sources
- Why this source over another? What did they hope to find, what did they take from it, and how will they use it?
- Corrections
- How did they catch it? What made them notice a mistake, why did it happen, and how did they fix it?
- Memory
- What stays the next day? With the work closed and no notes, what do they remember, and what do they now think they got wrong?
02 / What it’s for
Built for teams teaching AI to do real work.
Training computer-use agents
Recorded actions alongside the reasons people give for key decisions, including the options they weighed and the sources they chose.
Evaluating reasoning
A person’s account of a decision, made at the moment, to compare with what a model says and does in similar work.
Long-running work
How people plan, lose the thread, pick work back up and judge it later, across sessions and days.
03 / The people behind the work
Society runs on everyday people. AI should too.
Someone new to a task may stop at a step an experienced person takes for granted. An expert may recognize a pattern and know when a familiar approach will fail.
We ask both to explain it. We are building for a broad range of occupations and backgrounds, from people learning their way through a task to experienced practitioners and experts, each choosing their own work.
Each project includes the contributor’s actual résumé, with name and contact details removed, so you can see the experience behind the reasoning.
04 / The planned dataset
The explanation stays connected to the work.
See the full packageRecorded work
Screen video and narration, with keystrokes, clicks, scrolling, app context and session timing.
Thinking over time
What they knew at the start, decisions as they happen, pick-ups on later days and a next-day recall session.
Contributor context
The actual résumé alongside the project. Occupations and experience levels can be scoped to each collection.
Build the first collection with us
Start with your data need.
Tell us what you want to train or evaluate. We’ll scope a paid first collection around the backgrounds and data you need. An early review checkpoint lets you assess the first material against the brief.