LeetCode launches vibe-coding challenges: 12 projects, 60 minutes each, with an AI agent
LeetCode has launched Project Problems, vibe-coding tasks where users build a service together with an AI agent: they set the task, review the code, and fix errors. Each project gets 60 minutes in a sandbox, an isolated environment. The first cases include payment reconciliation like Stripe's and refund processing like Amazon's. In total, 12 tasks will be released over three weeks, so working with an agent can now be practiced the way algorithms are.
AI-processed from Habr AI; edited by Hamidun News
Classic LeetCode taught people to solve algorithmic puzzles: reverse a list, find a path in a graph, meet a required complexity bound. At work, engineers increasingly do something different: they explain to an AI agent what needs to be built, read its code, catch its mistakes, and take the result to production. Now the platform has added a format that tests exactly these skills: Project Problems, 12 project tasks at 60 minutes each.
What Project Problems is
Project Problems is a new task format on LeetCode. Instead of a short function with tests, the developer gets a small project to build with AI assistance. The format assesses the ability to manage AI agents: assigning them tasks, debugging the resulting code, and bringing development to a fully working product.
This is what people today call vibe coding: the person doesn't type every line but directs the process. Here the process has strict limits. The task runs in a sandbox (an isolated environment), and exactly 60 minutes are allotted. In that hour, getting neatly formatted code from the agent is not enough; you have to make sure it works.
The tasks mimic real work that engineers do at large IT companies. Among the first projects:
- a payment reconciliation system similar to Stripe's;
- an event recommendation system, as at Bloomberg;
- an inventory management API;
- a product returns processing mechanism, as at Amazon.
All four examples share a common trait: they contain a lot of business logic and edge cases. In payment reconciliation, amount mismatches and duplicate transactions almost always turn up, and in returns, partial cancellations and repeated requests. The agent will happily write the main scenario, but you will have to remind it of the awkward cases yourself. This is an example from general engineering practice; LeetCode did not disclose the specific tests.
Conditions and schedule
Users without a premium subscription get three attempts per day. From November 16 to December 6, 2026, LeetCode releases four new tasks every Monday, for 12 in total. Anyone who successfully completes at least 8 of them will receive a reward in the platform's internal currency, LeetCoin.
The reward is symbolic: it is the platform's own currency, and the material does not say it gives anything outside the platform. Most likely, this is a way to acquaint the audience with the new format and gather the first wave of participants.
Why it matters to developers and teams
For practitioners, there are two useful takeaways.
First: the ability to work with an agent is becoming a distinct skill that can be trained and measured. Previously, interviews checked whether a candidate remembered an algorithm. Now there is a venue that practices something else: how to phrase a task so the agent doesn't drift off course, how to quickly notice that code breaks on non-standard data, and when it is simpler to fix something by hand than to explain it again.
Second: the 60-minute limit reflects working reality well. An agent speeds up the start, but it does not cancel review. Someone who blindly accepts the first answer risks spending time debugging code they don't understand. Someone who breaks the project into small steps and checks each one usually gets further. This observation comes from practice working with AI assistants, not from LeetCode data.
Product managers and engineering leads will also find the news useful: it shows where the evaluation of developers is shifting. If a popular interview-prep platform includes agent management in its program, some employers will likely start asking about it in interviews too. This remains a prediction: LeetCode has not said the results will be used in hiring.
Limitations
The published information leaves the details of evaluation unclear: how a solution is scored, which tests are run, and whether one can choose a specific agent. It is also unknown whether the format will survive past the first three weeks or become a permanent section. For now, this is an experiment with 12 tasks, and it is too early to judge how close it comes to real work.
Still, the shift is noticeable: instead of asking "can you write this yourself," a popular platform asks "can you bring this to a working product together with an agent." Those who want to test themselves should try the first four tasks on November 16 and aim for 8 completed out of 12.
- Nov 21, 2026VibeCBS collects project nicknames in advance: the agent finds 2.8 times more documents
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026Sakana AI Opens Applied Research Lab for Market and Economic Modeling
- Nov 21, 2026Meta Is Reportedly Preparing the $300 Muse Charm AI Tamagotchi With a Year of Verizon 5G
Key Questions Answered
What Project Problems is?
Project Problems is a new task format on LeetCode. Instead of a short function with tests, the developer gets a small project to build with AI assistance. The format assesses the ability to manage AI agents: assigning them tasks, debugging the resulting code, and bringing development to a fully working product. This is what people today call vibe coding: the person doesn't type every line but directs the process. Here the process has strict limits.
What are the key details on Conditions and schedule?
Users without a premium subscription get three attempts per day. From November 16 to December 6, 2026, LeetCode releases four new tasks every Monday, for 12 in total. Anyone who successfully completes at least 8 of them will receive a reward in the platform's internal currency, LeetCoin. The reward is symbolic: it is the platform's own currency, and the material does not say it gives anything outside the platform.
Why it matters to developers and teams?
For practitioners, there are two useful takeaways. First: the ability to work with an agent is becoming a distinct skill that can be trained and measured. Previously, interviews checked whether a candidate remembered an algorithm. Now there is a venue that practices something else: how to phrase a task so the agent doesn't drift off course, how to quickly notice that code breaks on non-standard data, and when it is simpler to fix something by hand than to explain it again.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.