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.
If you ask AI to write posts but have to fix the tone and posting schedule every time, there is something to examine before the wording of your request: the brand guide the AI reads when it starts work. This video connects a brand folder, then uses a single question to check whether the brand description and prohibitions are accurately reflected in the answer.
Corner Coffee, the subject of the exercise, is a fictional brand, not an actual shop. Although the video was published on October 9, 2026, the folder connection and question demonstration are based on actual Codex screens from October 6. What was confirmed was the process of reading rules and answering, not an increase in sales or engagement.
If you want AI to follow the rules, first prepare accurate brand files and a question to check its understanding.
AI-generated staged image, not an actual work screen.
What belongs where in the folder?
The brand’s identity and tone go in context/brand.md, while the team’s working methods and prohibitions go in AGENTS.md. Actual records and reference material are collected in inputs, and outputs are organized into plans and reports under outputs. Separating source material from results makes it easier for a person to review what the AI referred to and what it created.
The parts of the brand file that need particular accuracy are channel-specific posting rules, tone, and photo and image policies. The Corner Coffee example specifies separate posting frequencies for Instagram and Threads and includes sample sentences for the desired tone. It also makes image-production authority explicit: the owner takes photos with a phone, and the AI only plans the shoots.
Opening hours and posting frequency are not fields where plausible examples should simply be copied. A single line that differs from actual operations can lead the AI to diligently plan around the wrong standard. That is why the current schedule must be checked against the file before drafting begins.
AI-generated staged scene 2View original
AI-generated staged image, not an actual work screen.
How do you check a claim that the files were read?
The rules file specifies that the brand file must be read before work begins, alongside boundaries such as not posting directly to SNS, producing drafts only, and not inventing reviews, discounts, or figures. It also requires the AI to state when information is unavailable, establishing a rule against filling gaps with imagination.
After connecting the folder, ask, “Tell me what our brand is and what you must not do in this project.” There are two things to check, rather than the length of the answer: whether the brand description matches the file and whether the answer actually states the prohibitions. If the response differs, revise the files and compare again using the same question.
Codex’s folder trust prompt should not be treated as permission merely to read. The video explains that it can read and modify files and execute code, and that some settings may allow code to run without asking. It is safer to grant trust only to folders whose contents you understand, and to check the scope of work before connecting them.
AI-generated staged scene 3View original
AI-generated staged image, not an actual work screen.
Why revisit even an accurate file?
The video also introduces records left in the folder of Seo-a, a fictional AI character. A strategy AI read guidance that differed from the actual posting frequency and scheduled too many Instagram posts. After the brand file was corrected to match the actual schedule, the posting dates in the new strategy changed as well. This illustrates how incorrect guidance can carry its errors into the output.
However, the earlier brand file was overwritten and no longer exists. The video cites wording preserved in strategy reports and planning documents, so this should not be treated as an experiment with complete before-and-after source files. It also notes that no actual business results for Corner Coffee were verified.
AI does not automatically know SNS posting history outside the connected folder. Actual records must be supplied as files, and an accurate answer about the brand does not validate all subsequent work. Before publication, a person still needs to check new plans and drafts against the schedule, facts, and prohibitions.
Watch this video to see the actual brand folder setup and check question
There is no need to build a large automation team from the start. Begin with a single page describing whom the brand serves, the posting rules for each channel, its tone, and its prohibitions. Connecting the files and asking the AI to repeat the brand description and boundaries is a small check that can help identify work starting from the wrong assumptions.
