Practice
A library of realistic problems in working codebases. Work at your own pace with AI.
- Fix production bugs from a support ticket
- Build features in apps you have never seen
- Review pull requests an AI wrote
- Ten-minute drills on one skill
Real engineering.Real-world practice.
Debug production bugs, build features, and review AI-generated code.
Work with AI in a real development environment.
Files
⌄shop-demo
⌄src
⌄api
TSwebhooks.ts
TSretry.ts
›cart
⌄payments
TScapture.ts
›models
›tests
›frontend
config
{ }package.json
MREADME.md
TS capture.ts ×
1export async function capturePayment(order: Order) {2 const idempotencyKey = `capture_${order.id}_${Date.now()}`;34 const existing = await db.payment.findUnique({5 where: { idempotencyKey },6 });7 if (existing) return existing; // already captured89 const result = await gateway.charge({10 amount: order.total,11 currency: order.currency,12 idempotencyKey,13 });1415 return db.payment.create({ data: { orderId: order.id, ...result } });16}TerminalTestsPreviewPorts
$ AI Assistant···
ChatPlanEditTools
Ask anything...
Report readyDouble charge at checkout
Your fix 18/18 Nothing else broke 325/325 Performance 9/9
One platform. Three ways to grow.
Practice real problems, simulate real interviews, or run AI-powered rounds for your team.
Interview practice
Timed rounds in the formats companies use for AI-allowed interviews. Coming soon: practice the same skills in the library today.
- AI-enabled coding round: fix, build, then make it fast (60 min)
- Bug bash in an unfamiliar codebase (45 min)
- Code review round (60 min)
For companies
Run realistic AI-era rounds for your candidates, graded on how they work, not just the answer. In design with early partners.
- Real codebases instead of puzzles
- Detailed reports on fixes, side effects and AI use
- Custom roles, rubrics and private questions
Problems
Pick one and start. No setup: the codebase, tests, preview and AI are ready in your browser.
- Real codebasesNot toy problems
- AI assistanceBring your own Claude or Codex
- Tests & previewRun and see changes
- Based on real workReal bugs and incidents
Graded on the fix, and on how you got there.
One report per attempt: what passed, what you broke, and how you worked with AI.
- Checked like real workSubmit runs a full test suite in a clean, separate environment. Passing locally is not enough.
- Collateral damage is caughtA fix that quietly breaks CSV export or another page fails the “nothing else broke” group.
- Then you explain it, without AIShort follow-up questions about your own change, the way an interviewer would ask them.
Invoice marked Paid after a partial payment
Example report Sample data, not a real attempt
Tests
Your fixThe bug is gone, checked in a clean environment
18/18Nothing else brokeExisting behavior still works: collateral damage caught
325/325PerformanceStays fast at realistic data sizes
9/9Four dimensions
Correctness and safety
Verification
Understanding
AI judgment and oversight
One thing to try next timeRun the existing invoice tests before asking the AI for a fix. You asked first, then checked.
Illustrative layout of a real report with sample data.
Free while we are in beta.
Every problem is open to you today. Paid plans are not on sale yet, and nothing here will charge you.
Free
Beta$0during beta
Everything that exists today.
- All problems: fixes, builds, reviews and drills
- Use your own AI: subscription login or your own API key
- Hidden-test grading and a skills report on every attempt
- Live workspace: editor, terminal, tests and preview
Pro
Coming soonTBAnot on sale
For people with an interview on the calendar.
- Timed company-format rounds
- Our AI included, no key needed
- Full reports with the follow-up questions graded
- Everything in Free
For companies
ContactTalkto us
Run AI-era rounds for your candidates.
- Real codebases, custom roles and rubrics
- Private questions for your team
- Reports on fixes, side effects and AI use
- Onboarding with the people building it
Beta prices are not final. See the full pricing details or write to hello@aicodinground.com.
Know the round before you sit it.
Short guides on how AI-era coding interviews work, with every fact linked to its source.
- Overview · 8 min read
What Is an AI Coding Round? How to Practice for One
What an AI coding round is, how much AI you get (chat-only or agent), what gets judged, which companies run one, and a 2-week plan to practice for it.
Read the guide - Comparison · 6 min read
LeetCode Alternatives in 2026: What to Use and When
LeetCode alternatives by interview round: why the AI coding round needs real-code practice, where to get it, and the other tools in one line each, with prices.
Read the guide - Skill · 7 min read
Debugging Interview Practice: Bug Bash in Unfamiliar Code
How debugging and bug bash rounds work, why your process counts more than a finished fix, a five-step loop for code you have never seen, and problems to practice on.
Read the guide
Questions, answered plainly.
One timed practice session. You get a ticket in a real multi-file codebase, work in a cloud workspace with an editor, terminal, tests and live preview, then submit. Rounds ask you to fix a bug, build a small feature, review AI-written code or write tests.
Mostly Python and TypeScript today, plus SQL and database problems. Java, Go, C# and React problems are being added. Each problem in the list shows its stack, so you only pick what you use.
Submit runs hidden tests in a fresh environment: one group for your fix, one for anything you might have broken, one for performance. The report then scores four dimensions: correctness and safety, verification, understanding (short follow-up questions you answer without AI) and AI judgment and oversight.
Yes, that is the point. Each round states what the AI may do. During the beta you bring your own AI: Claude Code or Codex run in the workspace terminal and sign in with your own account, or you can work with no AI at all. We grade how you use it, including whether you check its work, not whether you used it.
On easy problems it often can, and we grade whether you check its work. Medium and hard problems are modeled on real bugs and incidents and give symptoms and logs rather than hints. Understanding is graded separately from the hidden tests.
Realistic multi-file codebases built from the kinds of bugs that break production software. Every problem is verified: a known-good fix passes the hidden tests and the unsolved starting point fails them.
Nothing during the private beta. A paid Pro plan (timed company-format rounds, our AI included) is planned but not on sale, and we will tell you before anything changes.
Places are limited. While founding spots are open, anyone can sign up and start straight away. When they are full, join the waitlist and we invite people in the order they joined. People who are not invited within 6 months have their application deleted.
Yes. You can start a problem again as often as you like, within the daily beta limits. Each attempt gets its own report, so you can see what changed.
No. The formats on the interview practice page follow what candidates and the companies themselves have publicly described. Companies change their process, so treat them as practice, not a promise.
Practice on real code before the interview does.
Free during the beta. Founding spots are open: sign up with Google and start your first round today.