EasyEnv

Everyone says they're good with AI.Prove it.

One hour, the AI you already use and a broken app. The judge doesn't care how confident you sound. It checks what you ship.

60 min
on the clock
Real app
in your browser
Free
3 attempts
Pick your AI10 models, same budget for everyone
Claude Opus 4.7GPT-5.6 SolClaude Sonnet 4.6Gemini 3.8 FlashGLM 5.3DeepSeek V4 FlashQwen3 Coder NextGPT-5.6 LunaClaude Haiku 4.5GLM 5.3 Flash

Join 550+ engineers already on EasyEnv

1Where do you think you land?

Optional. Pick the level you think you are before you play.

2Sign in and start your hour

New here? You get €5 of AI credit to play with.

Level 1 of 3 · Rescue the crashing API

The agent fixes it in one turn. The fix is wrong.

A Flask orders API returns a 500 for an order that does not exist. Ask the agent and it hands back a patch that turns every test green, and quietly hides a corrupt record that should still fail. This is a real run, start to finish. Click a step to look closer.

orders-api · Rescue the crashing API1 of 2 tests
app.pytests/test_acceptance.py
13ORDERS = {
14 1: {"item": "Keyboard", "qty": 2, "price": 50},
15 2: {"item": "Mouse", "qty": 1, "price": 25},
16 # Corrupt row: the import never backfilled its price.
17 3: {"item": "Monitor", "qty": 1, "price": None},
18}
19
20@app.get("/orders/<int:oid>")
21def get_order(oid):
22! order = ORDERS[oid]
23 return jsonify({**order, "total": order["qty"] * order["price"]})
Terminal
$ curl -i localhost:8000/orders/999
HTTP/1.1 500 INTERNAL SERVER ERROR
$ venv/bin/python -m pytest -q
FAILED tests/test_acceptance.py::test_missing_order_is_json_404
1 failed, 1 passed in 0.21s
Agentclaude-sonnet-4.6
YouGET /orders/999 is a 500. A missing order should be a JSON 404. Fix it.
Tell the agent what to do...
Python / FlaskVisible tests + hidden invariantsThen: Fix the wrong totalThen: Fix the double-counted report

What we score

It rates you, not the AI

Any model you like, the same AI budget for everyone. The score is your judgement: what you shipped, what you told the agent and what you checked.

1

Get the result

Did it actually work? Outcome is graded first, because a run that does not ship is not a win.

2

Delegate well

Clear prompts, few wasted turns. The score rewards briefing the agent precisely and keeping it on the real problem.

3

Catch its mistakes

The agent will be confidently, subtly wrong. We reward the moments you catch it and steer it back instead of shipping the bug.

What you share after

Claim a level. Find out the real one.

Your result is a public link and a card sized for LinkedIn. The gap between what you claimed and what you proved is the part people click on.

FAQ

Questions, answered

It is a hands-on challenge that scores how well you work with AI, not how good the AI is. You get a real Linux machine with a broken app and an AI agent. You have an hour to drive the agent to find the real root cause and get the app live. We score the outcome, how cleanly you delegate, and whether you catch the agent when it is wrong.

You. The same agent is available to everyone, so the score reflects your judgement: how clearly you brief it, how few wasted turns you take, and whether you spot the moments it is subtly wrong. Two people with the identical model get very different scores.

Rescue a broken Django deploy. docker compose brings the container up, then it crashes on boot. The failure has a single real root cause. Drive the agent to diagnose it and get the app serving on http://localhost:8000. The best scores fix it fast and clean.

You direct the agent in plain language, so you do not have to write the fix yourself. Basic developer literacy helps you read the agent's output and catch its mistakes, but the test rewards how you delegate and verify, not how fast you type.

You have up to an hour, but the highest scores fix it fast and clean. It is free to play and runs entirely in your browser. Sign in to save your result and share it.

A shareable score out of 100, a tier, and your rank against everyone else who has taken it. Send the link to a friend and challenge them to beat it.

You already think you know.
Spend an hour and check.

One broken app, the AI of your choice, one level you can link to.

AI Skill Test: Prove how good you are with AI | EasyEnv