About MSSJ
한국어 →Not a company that knows AI.
A company that finishes with it.
We started in a small office in Yeongyang, a rural county in Korea. Deciding what to do and checking the result — those two ends stay in human hands. Everything repetitive in between is handed to AI agents. That is how we run education, industrial AX/DX, and agentic software development as one line, not a lecture that ends when the slides do.
Chapter 1 · The problem
Most AI training stops at “now you know.”
Founder Chihoon Shin holds a Ph.D. in computer science and spent ten years running national R&D programs at Korea’s Electronics and Telecommunications Research Institute (ETRI), with visiting-researcher stints at the University of Tokyo and Newcastle University.
He left the lab and set up in Yeongyang — a shrinking rural county in North Gyeongsang Province. While city AI courses stopped at teaching people how to use a tool, what the field actually needed wasn’t knowing — it was finishing.
That belief is baked into the name of our own methodology. MSSJ’s education engine, SimThink, reads as 深Think — deep thinking.
“AI’s essence isn’t speed. It’s depth.
Clear the noise, and think deeply.”
Chapter 2 · How we work
People judge. Agents execute.
This isn’t a pitch line — it’s how MSSJ runs day to day. Deciding what needs doing, and verifying what came back: those two ends stay firmly in human hands. Everything that repeats in between — gathering, analyzing, drafting, executing — goes to an agent.
Anything only a person can do — payments, signatures, contracts, anything touching personal data — a person still does, always. Which means you don’t need to become an AI expert. You hire the agent the way you’d buy an appliance: say what you need.
Published, not hidden
How this principle is enforced — which permission tier, which audit trail — isn’t kept behind closed doors.
See our operating trust pageChapter 3 · Measurement and scale
Measure small. Re-validate at organizational scale.
In July 2026, 25 civil servants in Yeongyang County spent four days building AI agents to take over pieces of their own jobs. Those who completed the hands-on track scored 93% on average. The working web apps they shipped grew from 7 in the first cohort to 13 in the next.
That same summer, ten elementary school kids met every Sunday for four weeks and made games by talking. By the end, 116 works were published, played and viewed by each other 1,415 times — we kept that story separately. Read the SimThink OS Story →
Outside the classroom, the same grammar is becoming products. A multilingual (5-language) AI health-consultation assistant, “Myeongsim,” is live at a local pharmacy. “Sabok-sabok,” an AI study companion for Korea’s senior social-worker licensing exam, is in beta. “Nongpani,” an AI tool for farmers selling produce direct, has finished design and awaits its next pilot. See the full project list →
A small site is not proof of scale. It is a place where the full path from request to completion can be seen and measured. MSSJ compares lead time, human touches, rework, and weekly throughput before and after a pilot.
A 10x candidate does not come from one generative-AI feature. It comes from removing repeated entry, running independent agent tasks in parallel, routing only exceptions to people, and closing the loop from planning through execution, quality, and records. Only workflows that show measured order-of-magnitude throughput or lead-time potential move toward productization.
Larger organizations add users, legacy ERP/MES integrations, permissions, security, and approval layers. We therefore productize the measured mechanism — data contracts, agent roles, exception handling, audit logs, and integration adapters — and validate it again under the larger organization’s conditions. System-of-record ownership, idempotent execution, human stop and manual takeover, staged rollback, and quality, safety, and compliance guardrails are fixed before expansion.
What travels is not Yeongyang’s result. It is the mechanism and measurement method that produced it.
Chapter 4 · Productization
Turn validated workflows into operating industrial systems.
Our education and field projects have produced a hands-on platform, agent execution pipelines, operating ledgers, and evaluation tools. We are now joining data connectors, agent orchestration, permissions and approvals, audit logs, evaluation, and monitoring into one operating stack.
On that shared stack, manufacturing gets Agentic MES and Smart Factory modules; engineering gets permitting and profitability-review workflows; public and service organizations get document, casework, and knowledge modules. We do not merely reskin screens by industry. We turn measured bottlenecks and exception rules into product specifications.
The goal is explicit — people own goals, exceptions, and approvals; agents carry repetitive execution.
MSSJ doesn’t sell lectures. People hold judgment, agents hold execution — and we transplant that same grammar into your organization.
Where these numbers come from
Every figure above comes from MSSJ’s own measured records of programs we ran ourselves. The full detail is published on each outcome page.
Next step
Myeongseong Simjae (MSSJ) · Yeongyang, North Gyeongsang, Korea · Founder Chihoon Shin, Ph.D. — End-to-End AI education, industrial AX/DX consulting, and agentic software.