About MSSJ
We help organizations finish real work
with AI agents.
Myeongseong Simjae (MSSJ) started in a small office in Yeongyang County, North Gyeongsang Province, Korea. People decide what to do and check the result. AI agents do the repeated work in between. We link training, industrial AX/DX (AI transformation and digital transformation) and development in one flow, so the work does not end with one lecture or one report.
Chapter 1. The problem
Most AI training stops once people know how a tool works.
CEO Chihoon Shin has a Ph.D. in Computer Science. He carried out government R&D for 10 years at the Electronics and Telecommunications Research Institute (ETRI). He was also a visiting researcher at the University of Tokyo and Newcastle University, where he followed research outside Korea.
After he left the lab, he settled in Yeongyang, a rural county with a shrinking population. AI courses in cities often ended with how to use a tool. What people in Yeongyang needed was help to finish their own work.
This idea is also in the name of our method. MSSJ's education engine is called SimThink. “Sim” is read from the character 深, which means deep. SimThink means deep thinking.
“The core of AI is depth rather than speed.
Clear away the noise and think deeply.”
Chapter 2. How we work
People make the decisions. Agents carry out the repeated work.
MSSJ works this way every day, inside our own company. People always keep two ends of the work: deciding what to do and checking the result. Agents do the repeated steps in between: collecting, analyzing, drafting and carrying out tasks.
Some work only a person can do, such as payments, signatures, contracts and handling personal data. A person always does this work. So you do not need hard training to use agents. You tell us what you need, as you would when you choose a home appliance.
We publish how we do this
We show which permission levels and which records we use to keep this rule.
How we keep AI reliableChapter 3. Measure, then scale
We measure in small places, then check again in large organizations.
In July 2026, 25 public officials of Yeongyang County spent four days building AI agents to do parts of their own work. People who completed the practice had an average score of 93%. The work web apps they made grew from 7 types to 13 types in the next group of trainees.
The same summer, ten elementary school children met every Sunday and made games by giving instructions in words. After four weeks, 116 works (games, videos and music) were on the class board. The children played or watched each other's works 1,415 times. We wrote this down as a separate story: Read the SimThink OS story
Outside the classroom, the same method is turning into products. At a local pharmacy, Myeongsimi, an AI that gives health counseling in 5 languages, is in real service. Sabok-sabok, for people preparing for the Level 1 Social Worker exam, runs as a beta. Nongpani, for farms that sell their produce directly, is waiting for its next field test. See all projects
A small site does not prove that a method will scale. What it gives us is a place where we can see the whole flow from request to completion. We can measure where people wait, type the same data again or go back a step. For one unit of work, MSSJ compares lead time (time from request to completion), human touchpoints, rework rate and weekly throughput before and after adoption.
One generative AI feature alone does not make work 10 times faster. We remove repeated data entry. Several agents handle independent tasks at the same time. Normal cases keep moving, and only exceptions go to a person. We connect planning, execution, quality and records in one closed loop. We build the workflow in modules so that it can be used again. Only tasks where we actually measure 10 times the throughput, or a 90% cut in lead time, become candidates for a product.
Large organizations have more users. Their existing ERP (enterprise resource planning) and MES (manufacturing execution system), permissions, security and approval steps are also more complex. So we do not copy the results of a small site. We turn the measured workflow into a product: data connection rules, agent roles, approval and exception handling, audit logs and adapters that connect to existing systems. Then we check it again under the conditions of the large organization. Before that, we fix these points: which system owns the master record for each type of data, how to stop the same task from running twice, how a person can stop the system at once or switch to manual work, how to roll back step by step, and the criteria for quality, safety and compliance.
We do not carry the Yeongyang results over to other organizations. We carry the mechanism that produced the effect and the way we measured it.
Chapter 4. Products
We turn tested workflows into industrial systems that can run every day.
In our training and field projects, we built our own practice platform, agent pipelines, operating records and evaluation tools, one by one. Now we are joining data connections, agent orchestration (coordinating many agents), permissions and approvals, audit logs, and evaluation and monitoring into one operating stack.
On this shared stack we add modules for each sector. For manufacturing: Agentic MES and Smart Factory. For engineering: permit and profitability review. For public and service organizations: modules for documents, citizen requests and knowledge work. For each sector, we turn the measured bottlenecks and exception rules into the product specification.
Our goal: people own goals, exceptions and approvals, and agents keep the repeated work going.
We bring the same way of working to your organization. People keep the decisions, and agents carry out the work.
Where these numbers come from
The class numbers on this page are measured records from training that MSSJ ran itself. The detailed records are published on each results page.
Next step
Myeongseong Simjae (MSSJ) · Yeongyang, North Gyeongsang Province, Korea · CEO Chihoon Shin, Ph.D. in Computer Science. End-to-end AI training, industrial AX/DX consulting and agent software.