The Planner Who Decides Where AI Goes, Not How It Is Built

One in five postings is not looking for a developer

On September 16, 2026, JobKorea published an analysis of the AI job postings placed on its platform in the first half of the year. Roughly 18,000 postings carrying an AI keyword were registered through June. Of those, the postings that companies themselves tagged as dedicated AI roles rose 128% from the same period a year earlier.

The interesting part is the composition. Postings JobKorea classified as non-developer AI roles rose 325% year over year, faster than the developer side. Non-developer roles went from 11.6% of all AI postings in the first half of last year to 21.6% this year. More than one in five is not a job building a model.

Within that group the ranking is clear. AI content creator grew fastest at 405%, followed by AI planner at 363%, AI business strategy at 313% and AI training consultant at 195%. JobKorea read this as demand spreading past the people who build AI toward the people who apply it in business, content and education.

Developer postings next door fell over the same months

Set the numbers beside developer postings and the direction sharpens. Over the same period, developer postings excluding AI roles fell 6.6% from the first half of last year. Web publishers fell the most at 40.1%, followed by app developers at 32.2% and game developers at 26.4%.

JobKorea read this less as AI replacing those roles directly and more as a rearrangement of the skills each role requires while AI hiring expands. A company representative said AI ability is settling in as a baseline across planning, business and content work rather than remaining the specialty of developer roles.

For anyone who plans products, the contrast carries one signal. The market is moving toward a place where knowing how to operate the tools is no longer what separates candidates. The three roles that shrank are all about building. The four that grew are all about deciding where the thing gets used.

The hiring process started asking first

The shift shows up more sharply in assessment stages than in job descriptions. On September 8, 2026, Kia announced a second-half hiring push covering 30 areas including production and manufacturing, IT, business and planning, corporate support, and research and development, open to both new graduates and experienced hires. Applications run from September 15 to 29 for new graduates and from September 23 to October 7 for experienced candidates.

For this graduate intake, Kia added an AI problem-solving assessment stage, which it describes as an industry first. According to the company, applicants are evaluated on problem-solving ability through the process of analyzing and solving a given problem using AI. Kia wrote that the intent is to hire people who can treat AI as a partner in solving problems, not simply people who can operate it.

JobKorea’s analysis shows the same pattern. Major companies including SK hynix and LG Electronics recently added items that probe AI ability in their hiring processes. Every applicant now meets that question regardless of the role they are applying for, and for planning roles the question is the job itself.

What this role actually decides

The success criterion here differs from the one on the model side. There the answer is performance; here it is whether attaching the model to a given task produced something better than the person who used to do it. So the first decision is scope. How many times a day does the task occur, can a wrong output be undone, and if it cannot, where does the human check go? Those belong on the first page of the planning document.

The second decision is how failure gets counted. A good outcome can be demonstrated, but the question this role has to answer is who pays when one attempt in ten is wrong. A feature shipped without deciding whether the customer absorbs it, the owning team absorbs it, or the company reverses it comes back once the support queue fills. Passing an assessment stage and then choosing where to actually apply the model are different skills, and the gap opens here.

The third decision is price. Model calls cost money per use, which makes it easy to build a price sheet that loses more as customers use the product more. How usage and price are bound together is the subject of a neighboring specialization in the same job family. In roles hired under the AI planner title, all three usually land on one person.

Getting in when the postings want experience

The entry conditions are already in the numbers. In the same analysis, experienced-hire postings for AI roles rose 140% while entry-level postings rose 55%. JobKorea read that as companies looking for people who can be deployed on real work immediately. Growth by company type also skewed large: 187% at major corporations including affiliates and subsidiaries, 167% at mid-sized firms, 124% at small firms and 117% at ventures. The door into this title straight out of school is narrow, and it widens with company size.

The realistic route is therefore to make these decisions inside the job you already hold. One documented case, showing which task the model was attached to, what improved, how often it was wrong and who paid when it was, carries more weight than a list of tools used. Kia’s assessment stage is described the same way: it watches the process of analyzing and solving a problem, not the ability to operate a tool.

This article rests on limited evidence, and the limits are worth naming. What JobKorea counted as a dedicated AI role, and what it classified as non-developer, could not be confirmed from published material. Posting counts are a proxy for demand rather than confirmed hires, and they cover one platform. A large growth rate and an equally large number of new seats are not the same fact.