Case study · AI baseline vs final

Seven weaknesses in the AI’s first version, and what changed

I gave a fresh AI agent the brief as written, with no design input, and kept what it made. The red rings show where each problem is.

1A leave time in the past still says “Soon”

The clock says 13:55, but the Up next card says “Leave by 13:53” with a “Soon” label. She’s already late, and nothing tells her.

AI baseline · 13:55
Baseline at 13:55. The Up next card says Leave by 13:53 with a Soon label

“Leave by 13:53”, two minutes ago.

Final · 16:02
Final fix screen. You're still at Belvédère. Leaving now, you'd arrive 16:18, 3 min late

She’s told she’s late and by how much, with ways to recover.

What changedWhen she’s late, the app says so, by how much, and what she can do.

2Free time shows three different numbers at once

All at 12:40, for the same afternoon.

Home
Baseline Home says 4h 15m free

4h 15m free

Plan form
Baseline plan form says a 5h 50m window

5h 50m window

Plan result
Baseline plan result says 1h 57m free time

1h 57m free time

What changedOne planner works out the free time, so every screen shows the same number.

3A full session gets one option and a score

The prompt asked for capacity changes. The “session full” alert pops up 5 seconds after she accepts a plan, whatever the time, and offers one other session with a “79%” score instead of a reason.

AI baseline
Baseline session full sheet with one other session labelled 79%

One option, with a score instead of a reason.

Final · the brief’s curveball, 14:10
Final fix screen. Session full, three other sessions to pick from, the best fit says why

Three sessions to pick from. The best one says why.

What changedInstead of one option with a score, she gets three sessions ranked by fit. Each says why it works (same topic, seats left, she still reaches Mateo on time). She can also watch it live or skip it. One tap swaps it, with Undo.

4Overlaps don’t say by how much

Adding a session that clashes shows a red “Overlaps” label with no times. She can’t tell what she’d miss.

AI baseline
Baseline conflict sheet with a red Overlaps label and no times

“Overlaps”, but by how much?

Final
Final conflict sheet. Overlaps The Jobs AI Will Create by 40 min, 14:30 to 15:10

“Overlaps by 40 min, 14:30–15:10”, and what she’d lose.

What changedIt shows the exact minutes and what she’d lose, then she decides.

5Made-up match scores

Every card shows a “% match” that nobody can check.

AI baseline
Baseline Discover list with percentage match labels

A “% match” on every card.

Final
Final proposal with plain reasons on each session

Plain reasons, like “AI and Energy, two of your interests”.

What changedNo percentages. Each session says why it’s there.

6Four different ways to edit one day

My day has an Edit button, “Re-plan rest of day”, “Add block”, and a ••• menu on every card. Too much to learn for someone with ten minutes between meetings.

AI baseline
Baseline My day with Edit, Re-plan rest of day, Add block and a menu on each card

Four controls that overlap.

Final
Final Calendar with a session's menu open, showing Swap session and Remove session

One ••• menu on each session, with Swap and Remove.

What changedOne way to edit. Swap shows the options, remove comes with Undo, and fixed meetings have no menu.

7A generic look

Navy app styling that could belong to any conference.

AI baseline
Baseline Today screen in generic navy styling

Could be any event.

Final
Final Today screen with soft Davos light, glass cards and a winter photo

Soft Davos light and winter photos.

What changedA calm look made for Davos, with one type and spacing scale.