On-Out, a Korean AI company, is positioning Agent F less as a reporting tool and more as software that can change where fashion inventory sits and when it moves.
According to Maeil Business Newspaper, Agent F is described as an execution system that ties together enterprise resource planning data, spreadsheets, logistics records and sales information. In fashion, that work includes sending initial stock to stores, flagging likely stockouts, recommending or setting store-to-store transfers, analyzing sales and supporting markdown timing. A generated report can sit unread; an allocation or transfer changes what customers can actually buy.
Maeil Business Newspaper reported an interview with On-Out chief executive Kim Seung-min that placed the product in planners’ day-to-day work: pulling figures from several systems, stitching spreadsheets and recalculating sell-through by product and store before acting. Agent F is meant to take over much of that assembly so a decision can still match a live sales signal. In fashion, a slow item may need a markdown while a fast size or color may need to move to a store that can sell it at full price; delay can turn an allocation problem into clearance.
On-Out has said the system is in use beyond demos. Fashion brand Challenger Golf applies automation on sales and inventory data for choices such as discount timing and stock run-down. Food company S&Food has linked production, inventory and sales, and Dongyang Technology has connected sales, quotations, installation, maintenance and reporting. The industries differ; the pattern is the same: leave systems of record in place and add an execution layer across them.
Fashion merchandising rules often live with experienced planners, not only in history tables. A 70% sell-through rate can mean different things by store tier, region, weeks since launch, category or whether replenishment is still possible. On-Out says it tries to capture those tacit rules with the raw data so the agent can follow criteria staff already use.
That raises a governance question: which decisions run automatically, which need approval, and which rules must hold even if a model sees a statistical opening. A transfer that looks good on a spreadsheet can add freight, starve another store or arrive too late. Authority to act should stay narrower than authority to recommend until a retailer has evidence the workflow is stable.
Public material does not offer a shared benchmark for decision accuracy, transfer acceptance or extra margin, and it does not show whether every Agent F workflow writes into ERP or some stay advisory. Those points decide how much operational power the product really has. The competitive loop in fashion remains sales signal, merchandising rule and inventory action—and how fast that loop can run without dropping the judgment that made the rule work.