Musinsa, one of South Korea's largest online fashion platforms, says it has rolled out AI coding tools across the company and raised developer productivity by about 75%, and now plans to extend AI beyond engineering — into content creation and customer service — as part of a company-wide "AI native" transformation. The number is striking. It is also the company's own, measured on an internal metric whose methodology Musinsa has not disclosed — which, against what independent research has found about AI and coding, is the part worth dwelling on.

The biggest reported gains came in software development. The number of developers using AI coding agents grew from 100 in a July 2025 proof-of-concept to roughly 700 by March 2026. Over that period, Musinsa says monthly output per developer rose from 57 to 99.5 on its internal productivity measure — a 74.7% increase — while the share of medium- to high-difficulty tasks climbed from 18.8% to 35.5%. The tool behind the push is OpenAI's Codex coding agent, which Musinsa's chief technology officer, Jun Jun-hee, has presented at OpenAI events as the core of the company's "AI native" workflow.

Musinsa frames the payoff in headcount terms. Without AI, it says, it would have needed to hire about 38 more people to produce the same output — an effect it claims it achieved at roughly 2.1% of the equivalent labor cost, with no additional hiring. It also says the time for new engineers to reach the performance level of existing staff fell from six months to three.

Why a Self-Reported 75% Deserves Scrutiny

Here is the catch, and it is not a small one. "Points per developer per month" is the kind of activity-based, self-defined metric that controlled studies have repeatedly found unreliable as a measure of real productivity. In a randomized trial published in July 2025, the research group METR had experienced open-source developers complete real tasks with and without AI tools — and found the developers using AI were about 19% slower, even though they believed they had been roughly 20% faster. The gap between how fast AI feels and how fast it actually is ran to nearly 40 percentage points.

Part of the problem is what gets counted. Vendor and internal productivity figures often track activity — commits, pull requests, lines of code — that naturally rises when AI is in the loop, without capturing the extra time spent reviewing and debugging what the AI produced. In one industry dataset, individual developers completed more tasks with AI but team review time jumped as pull-request volume nearly doubled. Survey evidence is similarly mixed: in Stack Overflow's 2025 developer survey, only about 16% of developers said AI improved their productivity to a great extent, while the largest group said it had little or no effect.

None of this proves Musinsa's number is wrong. A company that standardizes one coding agent across 700 engineers, trains people to use it, and measures consistently may well see genuine gains — and research does find AI tends to help more with onboarding and unfamiliar work than with deep work in mature codebases, which is consistent with Musinsa's claim that new-hire ramp-up time halved. The point is narrower: a self-defined metric with an undisclosed methodology is a reason to read "75%" as a directional, internal figure rather than an independently established fact.

The "AI Native" Roadmap and the Push Beyond Engineering

Musinsa is targeting a four-stage "AI native" roadmap it aims to complete by the first quarter of 2027, moving from initial adoption to full deployment and broadening AI use from a subset of engineers to all technical roles. The company has also leaned into the framing through hiring: its "AI native" developer recruitment drew about 2,000 applicants and placed 66 new engineers, with candidates evaluated mainly on how well they used AI tools to solve problems rather than on coding alone.

Beyond engineering, Musinsa says an internal content-generation tool, "AI Studio," is now used by more than 270 employees across over 10 departments, producing an average of more than 60,000 pieces of content a month, with productivity gains the company again puts as high as 75%. A "Visual Search" feature for image-based product discovery has been trained on more than 3 million fashion images. In customer service, Musinsa is testing an "AI CS Agent" in beta that handles post-purchase inquiries, combining large-language-model searches of standard procedures with FAQ lookups.

A Musinsa official said the company is accelerating AI adoption across customer experience, brand partnerships, and internal efficiency — citing recently introduced features such as an AI trend-curation service — and aims to build a competitive edge in fashion tech through AI. Whether the headline 75% holds up under an outside measure is something only Musinsa, with its own data, is currently in a position to show.

Frequently Asked Questions

Do AI coding tools actually make developers more productive?

The evidence is mixed. Some teams report large gains, but a 2025 randomized trial by METR found experienced developers were about 19% slower with AI while believing they were faster, and surveys show many developers report little or no effect. Results depend heavily on the task, the codebase, and how productivity is measured.

What is OpenAI Codex?

OpenAI Codex is an AI coding agent that can write, edit, and help execute software-development tasks from natural-language instructions. Musinsa adopted it as the core coding tool behind its "AI native" workflow, and its CTO has presented the company's use of it at OpenAI events.

How reliable is Musinsa's 75% productivity figure?

It is a company-reported number based on an internal "points per developer" metric whose methodology Musinsa has not disclosed. That makes it best read as a directional, self-defined gain rather than an independently verified measurement, since activity-based metrics can rise with AI use without capturing added review and debugging time.

What is an "AI native" transformation?

It refers to redesigning a company's tools, workflows, and even hiring around AI rather than adding AI to existing processes. Musinsa is pursuing a four-stage "AI native" roadmap it aims to finish by the first quarter of 2027, spanning engineering, content creation, and customer service.