Find it. Define it.
Research a real repository and pin the starting commit. Write an original, clear problem brief, then have a human refine and approve the expected behavior.
RESEARCH / ORIGINALITY / HUMAN APPROVALThe pursuit of the harder problem
Find the corner
it can't reach.
Villix is a quest to find the problems AI still gets wrong. Bring your technical judgment. Make the next challenge count.
For people who look a little closer.
Fig. 01 / The odd one8 × 7
Independent mindsHard, fair problemsHuman-verifiable answersShared progress
01 / The quest
The next leap starts
with a better question.
A convincing answer is easy to accept.
A useful challenge asks for proof.
Villix brings curious technical minds together around a shared pursuit: finding where AI's capabilities end, and helping move that boundary.
It takes an engineer who notices the missing case. A researcher who questions the result. A person who keeps looking after the answer seems obvious.
Find your place in the quest02 / How it works
Good questions.
Solid evidence.
The work is careful. The idea is simple.
Find a gap, prove it, make it useful.
Build the proof.
Create the tests, reference solution, and reproducible environment. Show that the baseline fails for the intended reason and the solution passes. Check quality, coverage, and fairness.
BRIEF / TESTS / SOLUTION / ENVIRONMENTTry it. Learn. Refine.
Run complete, unhinted agent batches. Inspect real solves and test gaps, repair the evidenced cause, and repeat the affected checks. Final review leads to a human submission.
AGENT TRIALS / REPAIR / REVIEW / HANDOFFA look inside the process
Step inside
the workroom.
Find a real problem. Write a clear brief. Build the tests and solution. Watch independent agents try it, inspect the evidence, and refine the work.
Watch the animated workflowVillix workroom8 chapters / Press play
One idea. A complete trail of evidence.
03 / Find your field
Different minds.
One shared pursuit.
Bring what you know deeply.
Stay curious about what you don't.
FIELD / 01
Software engineering
Real systems. Difficult bugs. The cases that a clean-looking patch can miss.
FIELD / 02
AI evaluation
Look past a plausible answer. Find out whether the reasoning holds.
FIELD / 03
Math & science
Follow the proof, question the assumption, and make every step count.
FIELD / 04
Agents & tooling
Test what happens when a model has to plan, use tools, and finish the work.
The principle
Hard enough to matter.
Clear enough to prove.
A challenge should reveal a limit through evidence.
That is what makes the work worth doing.
04 / Your next move
There's a place
for your curiosity.
You don't need to have every answer.
Start by asking a harder question.