The question is shifting from ‘Did you use AI?’ to ‘How do you work with AI?’

For job seekers, generative AI first entered hiring as a policy problem: whether it was acceptable in resumes, interview preparation or coding tests.

Recent 2026 hiring processes show a different use case. Nexon is explicitly giving candidates AI tools as part of problem solving, while KT says AI-use capability will be assessed even in operational, sales and marketing roles.

Two companies are not enough to define the Korean labor market. But they are useful evidence that some employers are moving beyond policing AI use and toward observing the quality of work produced with AI.

Nexon removed the traditional coding test from its game-programmer internship hiring

Nexon's official 2026 NEXONtorial for Game Programmer process runs from application screening to an AI interview, an AI-use competency assessment, role interview, team interview and entry. Applications close at 4 p.m. on September 7.

Electronic Times reported that the previous coding test was removed this year. Candidates instead use provided AI tools to solve a problem similar to real work, with the assessment looking at AI understanding, problem approach and structured thinking.

That does not mean coding ability has stopped mattering. A better interpretation is that when AI can participate in code production, the employer can widen the assessment to how a candidate decomposes a problem, chooses what to trust and validates the output.

KT is testing AI use outside AI-development roles

KT's graduate hiring covers network infrastructure operations, B2B consulting and sales, and B2C marketing and sales. These are not simply AI-research positions.

KT says role expertise and AI-use competency will both be important. Rather than testing AI trivia, the company says it will present job-related situations and focus on how candidates use AI in solving them.

This suggests AI use may be treated as part of ordinary work design in network operations, enterprise customer work and consumer marketing—not only as a specialist developer skill.

‘I use ChatGPT’ can quickly become a low-information statement

When a new workplace tool first appears, knowing the tool can itself be differentiating. As adoption spreads, the tool name tells employers less about the person using it.

The same may happen with ChatGPT, Claude, Copilot and prompt-engineering claims. If most applicants can list the same tools, hiring signals have to move deeper.

What becomes more informative is the work chain: what problem was being solved, what was delegated to AI, what was checked, what a human decided and what outcome followed.

Using more AI does not automatically mean stronger AI capability

Real work includes constraints that demos often hide: wrong answers, confidential data, copyright, security, customer responsibility, quality and safety.

A developer has to evaluate security, performance and maintainability beyond whether generated code runs. A salesperson cannot simply paste sensitive customer data into an external model. A network operator must understand the risk before applying an AI-suggested action to infrastructure.

Practical AI capability may therefore look less like maximum usage and more like knowing what to delegate, what not to delegate and where human verification remains mandatory.

Candidates need evidence of ‘problem → AI use → verification → outcome’

If employers care about application rather than AI trivia, preparation changes. Listing more AI tools or certificates does not by itself explain work capability.

A marketing candidate can provide stronger evidence by showing what information was classified, which errors appeared, what validation rule was added and how the research process improved. A developer can explain which draft AI produced, how tests or review caught errors and what changed in delivery time or quality.

AI experience becomes more credible when it is embedded in a work narrative rather than presented as a standalone technology label.

Employers also need to assess the process, not only the final artifact

When everyone has access to AI, a polished take-home result can make it harder to separate candidate capability from model capability. Yet banning AI completely can make an assessment less representative of the actual workplace.

Process evidence can help: what information the candidate selected, what they asked the model, where they became skeptical, how they verified the answer and what evidence supported the final decision.

Nexon and KT are not using identical assessment systems. The common signal is narrower: both are bringing the way candidates work with AI into the hiring conversation.

BANSEOG VIEW

When AI becomes common, hiring returns to evidence of judgment

There are two different ways AI can enter hiring. One is AI automating or supporting the evaluation of candidates. The other is employers evaluating how candidates themselves work with AI. The Nexon and KT examples are interesting mainly for the second reason.

If this expands, tool names on a resume will carry less information. Employers will care more about what work the person owned, where AI was used, what humans verified and what outcome was produced.

Employers face the same challenge. Instead of adding a vague requirement such as ‘strong AI skills,’ they need to define which decisions in the role can be delegated and which decisions remain human accountability.

As AI becomes ubiquitous, the scarce worker may not be the person who can use AI, but the person who can decide which AI output deserves trust and take responsibility for what happens next.

SOURCES

Primary sources and references

These two cases do not prove a market-wide shift or the superiority of a particular assessment design. The article interprets the common signal visible in the companies' disclosed hiring processes.