Course · AI-assisted coding round·1 chapter · 4 three-phase problemsNew format

The new round.
Direct the model, don't follow it.

You sit with a real codebase and an LLM at hand, and in 60 minutes you fix a bug, extend a feature, and optimize for scale. The skill being tested isn't how fast you can prompt. It's how critically you read what the AI returns, and how cleanly you defend the code you keep.

Level

Intermediate

Study time

14h

Lessons

Problems

Language

Java

Or scan the chapter list →

Package includes: Coding Interview Patterns + Low Level Design + AI Coding Interview

UrlShortener.java · phase 1 — fix the bug

Review AI output

YOU

Cache returns a stale slug after the underlying URL is updated. Patch resolve().

+ proposed patch4 changes

public String resolve(String slug) {
  String url = cache.get(slug);  cache.invalidateIfStale(slug);     // no such method  if (url == null) {
    url = store.lookup(slug);    cache.put(slug, url);             // NPE if lookup misses  }
  return url;  // claim: "O(1) amortized"          // true on hit only}

invalidateIfStale isn't on Cache. Hallucinated.

Cache put(slug, null) on miss, so the next read returns null silently.

Complexity claim only holds on a cache hit; on a miss it is O(lookup).

3 issues flagged · ready to revise·Phase 1 / 3

Open in editor

How it's taught

Real codebases.
Real AI output. Hidden tests.

Three-phase problems

Each problem is a multi-file Java codebase with a planted bug, a feature follow-up, and an optimization target, on the same 60-minute clock the round uses.

Read the AI, don't trust it

Every problem makes you read what the model proposes and decide what to keep: the hallucinated method, the wrong complexity, the silently broken logic.

Hidden behaviour tests

Use whichever AI tool you prefer: ChatGPT, Claude, Cursor, Copilot. A hidden test suite grades your final state across all three phases, not your prompts.

The 60-minute round

Three phases. One codebase.
Each grades a different thing.

Every problem in the course follows the same arc, so when the real round opens, the structure is already familiar.

01Read & fix0-20 min

Reproduce the bug

Read the failing test, locate the seam, and propose a fix. Interviewers grade the reading, not the typing.

02Extend20-40 min

Ship the feature

Add the new requirement on top of the now-passing code: boundaries, edge cases, and tests you write yourself.

03Optimize40-60 min

Make it scale

Name the bottleneck, choose the trade-off, defend the complexity. Often where the AI is most confident and most wrong.

What you'll learn

The skills the round
actually grades.

Interviewers say the same thing: candidates who paste the first AI answer fail, and candidates who can read, scope, and direct the model pass. This course gets you to the second group.

How the AI-assisted coding round actually works at Meta, Shopify, Canva, and similar companies.

The three-phase arc (fix a bug, extend the feature, optimize for scale) within a 60-minute slot.

Reading AI output critically: spotting hallucinated APIs, off-by-one bugs, and wrong complexity bounds.

The reproduce, isolate, hypothesize, fix, verify loop, applied to multi-file codebases.

When to consult the LLM vs. when to think alone. Prompt scoping, context-building, and verification.

Common candidate failure modes: pasting unreviewed AI output, skipping the reading phase, accepting the first answer.

Hands-on practice with real interview-shaped problems where you fix bugs, add features, and optimize, using whichever AI tool you prefer.

Curriculum

Everything in the course.
Click any chapter to expand.

1 chapter·9 lessons

CConceptPProblemSSolution

The problem set

Interview-shaped problems, end to end.

Each problem ships with the starter repo, the planted bug, the feature spec, the optimization target, and a hidden test harness for all three phases.

01Medium

Cart Discount Engine: Fix, Extend, Optimize

Event Log Service: Fix, Filter, Tail

Word Counter: Fix, Configure, Rank

Bank Ledger Bug Hunt: Five Production Bugs

Where the format is showing up

Companies like Meta, Shopify, and Canva have swapped a classic coding round for this AI-assisted format, with several others piloting it.

Three courses. One unlock.
Coding, Low Level Design, and AI Coding Interview.

The pack covers Coding Interview Patterns, Low Level Design, and the AI Coding Interview course end to end: every concept, every problem, and every hidden test.

AI Coding Interview

The new round.Direct the model, don't follow it.

Real codebases.Real AI output. Hidden tests.