Practice working with AI

Courses tell you about AI. Tests quiz you. Simbox is where you practice.

A live queue, real files, an assistant that's helpful but fallible, and a deadline.

Guided demo · sample data · no account

Confidently Wrong · 5 min 4:32
Tasks
EMAIL · 09:02 Dana Okafor · Finance Q3 variance summary before standup.
CHAT · 09:04 Priya · manager Sanity-check whatever the AI gives you.
Assistant
YouSummarize Q3 spend variance. Flag anything over 5%.
Assistant
j/k tasks · ⌘↵ submitRecorded

The subtotal doesn't sum. 61% of people send it anyway.

Fig. 01 — A live session. New tasks arrive while you work.


"Proficient with AI" is self-reported.
Simbox watches you work.

Courses

Teach about AI, certify completion.

Measures watching

Tests

Multiple choice — completable with AI.

Measures trivia

Simbox

A simulated workday, fully observed.

Measures behavior

Practice. Review. Improve.

EMAIL Board summary, 10:30 sharp
CHAT "The spreadsheet has duplicates."
NEW Board moved up 15 minutes
ASSISTANT "Q2 totaled 48,210…"
Step 1

Work a real morning

Live tasks, colleagues who push back, a deadline that moves.

Assistant"48,210 visits. Nothing odd." You sent"47,880 — duplicates removed." The source47,880
Step 2

Read your review

Key moments, quoted word for word from your own session.

Step 3

Build a skills profile

Eight skills, moved only by evidence. They fade without practice.


After every session

A review, not a grade.

What the assistant said, what you sent, what the source actually says — and where your session landed against everyone else's.

Open a full review

Fig. 02 — Evidence, never vibes.


Eight skills, measured from behavior.

People are bad at judging their own AI skills, so Simbox never asks.

Fig. 03 — One person's profile, scored 0–4. There is deliberately no single "AI score".


Levels are earned, not ground out.

Every level needs demonstrated skill plus a passed assessment — never points or time in the app.

BeginnerEnrolled
Developing1.5+ in three skills
Capable2.5+ in five skills · one assessment passed
Advanced3.0+ in all eight · a harder assessment
Expert3.5+ in all eight, recently practiced
Fig. 04 — Requirements are published in full. Skills fade without practice, and your profile says so.

Different every time. Measured the same.

Names, numbers, and context regenerate for every attempt — only the scoring stays fixed.

Your attempt

Maya Chen needs the Q2 visits summary by 10:30. Total: 48,210 — one planted error.

Anyone else's

Priya Nair needs the Q3 claims summary by 15:00. Total: 31,540 — one planted error.

Both scored on exactly

Checked the total against the source · labeled rough numbers as rough · verified before sending

Fig. 05 — Nothing to copy, share, or find online. Easy to build, hard to game.


One product, three ways in.

Same scenario, same scoring — differing only in who sees the result.

Individuals

Train yourself

A five-minute challenge at lunch, a longer session weekly, an assessment when you're ready.

Daily #214 · 4:32
Streak 12 days · ready for Level 3
Companies

Upskill, provably

Baseline, six weeks of practice, re-measure. The answer to "did it work?" is observed behavior.

Team × eight skills · +0.6 verification
For companies →
Recruiters

Place with proof

Send a verified skills profile instead of "trust me, she's good with AI."

M. Reyes · Level 3 · Capable
Verification 3.2 · consent-gated
For recruiters →
No videos · no slides · no multiple choice

See it for yourself.

A guided walkthrough with sample data — the daily challenge, a full session, and the review it ends with.

See the demo