[Edaily Reporter An Yu-ri ] KRAFTON(259960)is expanding its “AI First” strategy to the recruitment sector. Following its announcement last October that it would transition into an “AI First” company, the firm has developed and implemented its own recruitment and evaluation solution that analyzes the entire process of applicants collaborating with AI.

According to the IT industry on the 28th, KRAFTON has developed and is implementing an AI-based recruitment and evaluation solution called “Cofa-Probe.” Cofa-Probe is an AI-native recruitment and evaluation solution that assesses the process by which candidates work with AI to define problems and verify results. It virtually recreates an environment similar to the actual work of an AI engineer and evaluates the process by which candidates interact with various Agent AI agents to identify and solve problems.

Unlike traditional evaluations that focus solely on the final output, Cofa-Probe structures the entire process of working with AI—from problem definition, assigning roles and issuing instructions to the AI, and searching for evidence, to revising judgments, recovering from errors, and verifying results. The AI provides the basis for evaluation and generates reports. KRAFTON utilized OpenAI’s Codex in COPA Probe. By applying a “meta-harness” that repeatedly tests and refines the LLM-based evaluation system, the company enhanced both the consistency and discriminative power of the evaluation.

Candidates collaborate with a task agent that reflects real-world roles, task contexts, and constraints. COPA Probe analyzes changes in questions raised during the task, as well as evidence-seeking, decision-making, iterative improvements, and verification records, and compiles this into an evaluation report.

Copa Probe evaluates AI-native competencies along three axes: Technique, Intent, and Cognition. It assesses the executability, structure, and security of the candidate’s code and deliverables; the methods used to convey objectives and constraints to the AI and verify tasks; and the candidate’s ability to explain why their deliverables function as they do and where they might fail.

During the evaluation process, the system automatically collects AI collaboration records—including source code, conversation logs, prompts, tool configurations, and verification execution traces. After the task is completed, a post-task Q&A session is conducted where the candidate explains their results and decision-making process, and a report is generated based on competency-specific evaluation criteria for use in automated scoring and as a reference for interviews. Copa Probe helps organizations recruit AI-native talent by structuring the basis for evaluation throughout the candidate’s workflow and providing consistent reports.

Copa Probe was also utilized in a recruitment-linked “hackathon” jointly hosted by KRAFTON and CJ Corp. on July 30. Based on the operational experience and on-site feedback gained from the “Copatthon,” KRAFTON plans to further refine Copa Probe’s evaluation criteria and agent environment.

Park Jae-min, Head of KRAFTON’s AI Frontier Division, stated, “Competitiveness in the AI era stems from defining problems alongside AI, verifying evidence, and validating results through to the end.” He added, “Cofathon was an opportunity to observe participants’ actual work methods beyond their final deliverables, and KRAFTON will further refine Cofa-Probe to establish new standards for evaluating AI-native talent.”

Kim Tae-young, Account Director at OpenAI Korea, remarked, “This collaboration is particularly significant because Cofathon was used not only to demonstrate how Codex supports actual work but also to validate and refine the Cofa-Probe evaluation system.” “OpenAI will continue its collaboration with KRAFTON and work together to explore various possibilities so that OpenAI technologies, including Codex, can be utilized more deeply in real-world work and in solving organizational problems,” he said.