In October 2026, we added new questions to the AI Collaboration format.

This release includes questions for roles where writing code is not the main job: product marketing managers (PMM), pre-sales staff and project leads at systems integrators, and security specialists. We also added a question for machine learning engineers, a practical development question for interns and junior engineers, and a C / C++ / Python version of the meeting room algorithm question.

You can add any of them from the AI Collaboration question list on the test creation screen.

Questions added

Question Theme Question Mode Screen Layout Difficulty Expected Time Limit
Interviewing customer success and prioritising feature requests Other Conversational AI chat only Easy 20 min
Interviewing engineering and briefing customer support on a slipped release Other Conversational AI chat only Normal 30 min
Reading an RFP: bid decision and clarification questions (pre-sales) Other Conversational AI chat only Normal 55 min
Change request or original scope: negotiating with the client Other Conversational AI chat only Normal 60 min
Security review of partner document upload into an on-premises business system Other Conversational Editor Normal 60 min
Repairing an image classifier that underperforms in production Other Non-conversational Editor Normal 120 min
Fixing Bugs in a Lesson Reservation System (6 variants) Practical Development Non-conversational Editor Normal 90 min
Compute the minimum number of meeting rooms from a reservation log in C, C++, or Python (clarify the rules via dialogue) Algorithm Conversational Editor Normal 45 min

Theme, Question Mode, and Screen Layout are filters in the question list.

Questions for roles that don't write code

In these questions, candidates gather facts by questioning an AI acting as a stakeholder, then put the result into a single Markdown file. The situation is not in the problem statement and can be learned only through the conversation. The questions check whether candidates dig past the first answer and commit to a decision based on the facts they confirmed. No coding is involved.

Questions set to "AI chat only" are taken on the candidate screen built around the AI chat. The file tree and terminal are not shown.

Questions for PMMs

  • Interviewing customer success and prioritising feature requests: As the PMM of a direct-booking service for hotels and inns, the candidate asks an AI playing the customer success (CS) lead about four feature requests and decides which requirements go into the next development slot. It checks whether they draw out the facts needed for the decision, set priorities based on those facts, and communicate the decision in a form that development and CS can act on.
  • Interviewing engineering and briefing customer support on a slipped release: When a new feature's release slips, the candidate checks the situation with an AI playing the engineering PdM and writes a briefing for the CS team, who will explain it to the hotels. It checks whether they verify assumptions instead of taking them for granted, separate what may be promised to hotels from what may not, and put the engineering situation into words a hotel will understand.

Questions for systems integrators

  • Reading an RFP: bid decision and clarification questions (pre-sales): The candidate reads an RFP for replacing a manufacturer's production management system, together with the company's own information, and writes a bid recommendation and clarification questions for the customer. With the RFP's terms out of line with the company's prices and track record on lead times, it checks whether they can turn vague demands into ten or fewer questions and stated estimate preconditions, draw out the customer's circumstances from an AI playing the account sales rep, and state the bid decision with its basis.
  • Change request or original scope: negotiating with the client: During integration testing of a contracted development, the client presses that "this report was in scope from the start." As the project lead, the candidate writes a response plan and a reply. It checks whether they can judge the scope from the contract, requirements definition, and meeting minutes, estimate the impact on effort, schedule, and quality, draw out the client's real constraints from an AI playing the client, and present options and a reply that hold the line without straining the relationship.

Question for security specialists

  • Security review of partner document upload into an on-premises business system: As a security specialist in the Information Systems Department, the candidate reviews a proposal to add partner document upload to an on-premises screening system. The proposal says nothing about how it is implemented or operated, so the candidate cannot judge what could happen without questioning the AI playing its designer. It checks whether they dig into the realities, such as when the virus scan runs, where the server sits, shared accounts, logging, and the maintenance vendor's connection, and then commit to a priority order for what must be done before launch and what to set aside within the go-live date and budget. This question is taken on the editor screen.

Machine learning question

Repairing an image classifier that underperforms in production

A bird species classification pipeline reports validation accuracy of around 0.9, but gets only about half right in production. The cause is not the model's capacity but somewhere in the pipeline. Besides making production predict correctly, candidates must make the validation accuracy the pipeline reports trustworthy as an estimate of production accuracy.

Candidates implement in PyTorch and can use a GPU in the test environment. It is aimed at machine learning engineers with hands-on PyTorch experience.

Practical development question

Fixing Bugs in a Lesson Reservation System

In a fitness gym's lesson reservation system (an API server and a daily job), candidates investigate and fix bugs in the "transfer" feature and the daily job (sending reminders and handling no-shows) that a predecessor added, and write tests to keep them from coming back. There are only two reports, but the same system holds ten more bugs nobody has noticed yet. The question evaluates whether candidates go beyond the reported symptoms and follow the business flow to find bugs of the same kind.

Available in six variants: Django / Express TypeScript / Go / Laravel / Ruby on Rails / Java / Spring Boot. It is aimed at candidates for IT engineering internships and junior engineers with about one to two years of experience.

Algorithm question

Compute the minimum number of meeting rooms from a reservation log in C, C++, or Python (clarify the rules via dialogue)

This is a version of "Compute the minimum number of meeting rooms from a reservation log (clarify the rules via dialogue)," added in August 2026, with the supported languages narrowed to C / C++ / Python. Candidates pick one of the three. Choosing C or C++ means building the log reading and memory handling by hand.

The log specification is not in the problem statement; candidates confirm it by asking an AI acting as the interviewer. It suits hiring for embedded and systems-level positions.

How to use them

Add any of these questions from the AI Collaboration question list on the test creation screen. You can filter the question list by Theme, Question Mode, and Screen Layout.