This is a republication of Reedo’s Naver article. It is not a new experiment or a DMS client case study. The cover and in-article photographs are AI-generated staged scenes, not evidence of actual work.
When entrusting strategy to AI, one of your biggest concerns may be plausible-looking performance figures that were simply made up. This video shows the actual process of creating a growth strategist and a content writer using four-part job descriptions, and having missing metrics reported as “no data.”
Corner Coffee is a fictional brand, and the performance table is empty practice material. Although the video was published on October 13, 2026, the Codex screens showing the agents being created and called come from a demonstration on October 6. It is important to distinguish the process of defining two employees from the scope of the work actually carried out.
Separate role, personality, tasks, and rules, and have missing figures left as “no data.”
AI-generated staged image, not an actual work screen.
How does a skill differ from the person responsible?
In the video, a skill is a work manual for repeating a set sequence and format, while an agent is the person responsible for deciding when to use that manual and what to do first. Rather than just giving it an impressive name, fill in four sections, role, personality, tasks, and rules, to document the decisions you expect and the boundaries it must respect.
The first employee, the growth strategist, is responsible for setting the brand's growth direction. Its personality is described as calm and realistic, favoring actionable ideas over flashy proposals. A condition against citing numbers without evidence is added, specifying not just its tone but also its approach to judgment.
Its tasks are to read the brand files and records in inputs to understand the current situation, select one to three priorities, and save them in a short strategy document. If daily content is needed, it should select and use the existing daily content set skill. This structure specifies both the reference materials and the tools available to the employee.
AI-generated staged scene 2View original
AI-generated staged image, not an actual work screen.
How should missing figures be handled?
The rules explicitly require it to write “no data” rather than invent follower or view counts when those figures are unavailable. The data principles in the resulting definition file also say not to treat empty values as zero. An unmeasured value and an actual zero are different facts; confusing them undermines the strategy from the start.
Another boundary prohibits using fictional records as evidence of actual customer behavior or performance. The metrics table in the practice folder was empty, so there was no reason to fill in numbers just to make a report look good. Separate actions such as generating images, posting directly to social media, and contacting outside parties are also distinguished from the permitted scope.
Immediately after the growth strategist was created, only its format and path were checked; the strategy task itself had not yet been run. Later, the creator called it by name and requested an operating strategy for October 8–14 and a content set for the first day. The actual reply reported that two files had been saved and explained that no growth effect was predicted because performance figures were unavailable.
AI-generated staged scene 3View original
AI-generated staged image, not an actual work screen.
Did both employees actually perform work?
The second employee, the content writer, is created using the same four sections. Its role is to produce a day's content based on the strategist's direction, first reading the latest strategy and posting records, then writing a set for the requested date. Its personality is defined as a meticulous writer who understands the brand's voice and mood, and it must also follow the prohibitions in the brand documents.
Do not interpret the scope of execution more broadly at this point. The video explicitly states that it only went as far as creating the content writer, and that the writer had not yet actually written any content. Examples of how to call it with a name, date, and mood do not verify a finished result produced by that writer.
The video also presents an example in which the strategist for Seo-a, a fictional AI character, recorded follower and view counts as “no data.” This does not establish the approach's growth effects or the quality of its strategic priorities. Neither an increase in actual response to Corner Coffee nor the quality of its strategy was verified, and a person must review the strategy and writing before using the results.
Watch how the employees were created with four sections and the strategist was called in this video
Start by defining just one AI employee in four sections. Write down what it should judge, which materials it should read, and how far it is allowed to act, and include a rule for leaving missing figures unfilled. Then call it by name and assign a small task. Check separately whether a file was created and whether its contents are trustworthy.
