MUSINSA announced one developer job family; its openings still differ

On November 18, 2026, South Korea's MUSINSA announced it would bring formerly separate server, web and mobile developers into a Product Engineer (PE) job family. The company is hiring experienced engineers and using AI agents in practical assessments. Candidates' solutions and their decision-making process are reviewed, with the final hiring decision made by human engineers and recruiters.

Reading the actual openings reveals different schedules. Service Platform specifies a 90-minute AI Agent Test, while Retention specifies 120 minutes. Both add a 30-minute review period and a 60-minute AI Deep Dive on the same day.

A third position, Core AI, still specifies live coding or architecture design at the first interview. An applicant should not assume that one unified PE job title means all engineering interviews now use the identical test.

Specialist requirements remain: seven, ten and eight years

Service Platform asks for at least seven years of engineering experience, Java or Kotlin with Spring, REST or gRPC interface work and operational experience with distributed systems. Its work covers shared backend APIs, developer portals, platform standards, messaging and large-scale traffic.

Retention asks for at least ten years of backend experience and specifies Spring Boot, Kafka, RDBMS and JPA or QueryDSL, Redis, distributed locks and incident management. Work spans customer retention products, rewards and event-driven services where consistency and peak-demand performance matter.

Core AI asks for eight years or comparable experience, with Java/Spring/JPA, database design and AI coding agent usage. Its remit includes evaluating third-party AI systems and building commerce-facing AI applications. The organizational label is shared, but technical specialization remains explicit.

The same-day test schedule is three hours or three and a half

Service Platform's 90-minute AI Agent Test is an in-office architecture-design assignment based on plausible fashion-platform functions or products. Candidates may use the supplied Codex and Claude tools. It is followed by 30 minutes to review the work and a 60-minute discussion—three hours in the listed schedule.

Retention's architecture exercise lasts 120 minutes and has the same 30-minute review and 60-minute Deep Dive, totaling three and a half hours. MUSINSA says it provides the MacBook and AI environment, with no need for candidates to prepare a separate account or device.

The job descriptions do not explain why one team allows half an hour more. Neither greater difficulty nor a higher score threshold can be inferred from time allocation alone.

Candidates cannot rewrite the submission during the review break

After the exercise, candidates can review their output and the prompts exchanged with the AI agent. They cannot edit the submission during the 30-minute break. The next interview requires them to reconstruct why they made particular choices.

The 60-minute AI Deep Dive covers problem framing, architectural trade-offs, what candidates delegated to AI, which suggestions they accepted or rejected, and how they verified the proposed solution. Interviewers discuss the artifact already submitted.

For example, proposing Kafka for order processing invites questions about duplicate messages, failure recovery and why asynchronous messaging was preferable to a synchronous API. This is an illustrative engineering question, not a disclosed MUSINSA exam item. The actual prompts and scoring weights have not been published.

Core AI's job posting preserves a distinct selection route

The Core AI job listing describes an initial interview involving live coding, with architecture design possible depending on the process. A later interview and compensation discussion follow. The listing also asks for hands-on AI agent experience.

Knowing how to use an AI coding tool is not the same as being assessed through the newly published AI Agent Test. The AI Agent label explicitly appears on the Service Platform and Retention openings reviewed for this article.

Applicants should inspect the relevant opening at the time they apply, not infer a permanent company-wide examination policy from a press announcement. These are the published conditions for specific recruitment tracks.

MUSINSA had already tested AI skills in junior recruitment

On March 30, MUSINSA announced that 66 people had been selected and assigned to real projects through its MUKIEs junior developer recruitment. Around 2,000 applied; approximately 1,700 took an initial coding test, about 400 participated in an AI-usage test and about 100 reached an in-person interview.

The company described a six-month period working on commerce projects followed by a regular-employee conversion assessment based on each person's capabilities and contribution. The announcement does not prove that all 66 subsequently converted to permanent status.

The November senior openings emphasize documented architectural choices, operational incidents, data integrity and measured product effects. There is a clear progression from assessing AI use among junior candidates to using AI-assisted exercises for experienced hiring, although there is no evidence that the tests or scorecards are identical.

Prepare one defensible problem-solving story, not just a stack list

The job descriptions require a résumé and make a portfolio optional. The portfolio guidance asks candidates to state the problem, the strategy used, personal contribution and observed outcomes. For AI-assisted work it asks how the tools were used and how the output was checked.

Consider a developer who reduced duplicate charges in checkout. A convincing account would identify the triggering conditions, rejected alternatives, the chosen idempotency controls, tests and post-release metrics. A list of frameworks alone would miss the type of judgment MUSINSA says it wants to discuss.

Applicants should be ready to distinguish work delegated to an AI agent from design decisions, safety checks and verification they personally owned. This example is preparation advice based on the posting, not a prediction of a real exam question.

The effect of the assessment itself remains unproven

MUSINSA is extending engineers' stated responsibility from implementation into user-problem definition, prioritization and post-launch measurement. Its AI-assisted evaluation explicitly asks the candidate to defend rather than merely submit generated work.

That does not establish that the new process predicts performance or is fairer than live coding. The company has not released evidence comparing subsequent performance, selection accuracy or bias across these assessments.

For now, the useful public evidence is the team-specific technology requirements and exact schedules. Candidates and employers can evaluate the reasoning behind the approach without treating its long-term results as settled.

BANSEOG VIEW | One job family, multiple tests

Service Platform and Retention both use AI Agent Tests, but they have different specialist requirements and 90- versus 120-minute exercises. Core AI retains a live coding or architecture option. Title alignment is not examination standardization.

A candidate should be able to explain problem selection, technical trade-offs, what the agent contributed and what they personally validated. MUSINSA gives candidates 30 minutes to review but not alter their submission before the Deep Dive.

An employer considering AI-assisted evaluation should distinguish the quality of an artifact from the engineer's own judgment. MUSINSA's method is observable; its hiring outcomes still need evidence.

Primary sources and references

Information checked November 19, 2026 against MUSINSA-authored LinkedIn job descriptions, its March 30 corporate release and November 18 Korean coverage. Test times are specific to Service Platform and Retention; Core AI has a distinct assessment description. No public evidence establishes difficulty differences, scoring criteria or post-hire performance outcomes. The 66 junior developers were selected and assigned to projects, not confirmed to have all completed permanent conversion.