The change Wallie introduces is grammatical before it is technical. For a decade, fashion’s inventory software has produced nouns: a forecast, a report, a dashboard a planner consults on Monday and overrides by Thursday. WAIR, the Amsterdam company behind it, has built an agent that produces a verb. Wallie sets opening quantities, halts replenishment to a store that has stopped selling, and moves units toward live demand ahead of a markdown, none of it waiting for a person to approve the step. The shift from advising a decision to making one is the whole story, and it is fashion’s first real test of delegation: whether an operator will let a model act on its forecast, or only read it.

Inventory is the right place to run that test, because it is where fashion’s margin is won and lost in small increments. WAIR’s chief executive, Mitch van Deursen, frames the prize in leverage: he claims that a 1 percent revenue gain from better inventory can lift bottom-line profit by 12 percent. The figure comes from the company’s own promotional content and should be read as a pitch, but the direction is sound. Markdowns, overstock, and unmet demand convert almost directly into lost margin, which is why a day shaved off each stock decision is worth more here than the same speed applied to a product description. An agent that acts on inventory therefore aims at the line of the income statement retailers watch most closely.

The forecast is the commodity; the authority to act on it is the product. Accurate demand prediction is now a crowded claim, and WAIR’s ForecastGPT2.5 model sits among a dozen rivals quoting similar numbers. FashionUnited’s billing of Wallie as “the first AI Fashion Agent for inventory” is partner content, not a verified milestone. What is new is that Wallie closes the loop: it enacts a decision first and reports after, while sibling agents named Whoopie and Suzie handle the planning numbers and the product copy. Its real effect is to delete the pause in which a person used to say yes.

The market, meanwhile, is asking for the opposite of what Wallie is built to do. In RELEX Solutions’ January 2026 survey of 514 retail and supply-chain leaders, 54 percent said they want AI to recommend while a human keeps the final call. That is a precise description of a dashboard: analysis on tap, authority withheld. Only one in ten would trust a system to decide on its own, a split DC Velocity glossed as trusting AI for support but not to act. The gap between those two figures is the gap between an insight and an instruction, and it is the one Wallie is built to close.


For now, autonomous inventory is more claim than practice, even in the research that promotes it. A November 2025 framework for agentic replenishment, presented at a conference in Bahrain, was tested on artificial data tables at a single mid-sized store, not across a live chain. Corporate practice is not much further along: McKinsey’s 2026 survey of roughly 500 organizations found only 23 percent had scaled AI agents somewhere in the business. The barrier they named most often was security and risk, and McKinsey states the fear plainly: the danger has moved from a system that says the wrong thing to one that does it. A wrong recommendation dies on a screen; a wrong action ships to a store.

The strongest objection is that fashion has already delegated inventory decisions, so Wallie renames a threshold it never crossed. Automated replenishment rules have moved stock without a human for years, and Lojas Renner proved at scale that handing allocation to a machine can grow revenue on less stock. But Renner delegated to a system it built and spent six years learning to trust; Wallie asks a retailer to trust an agent that arrives configured out of the box. A replenishment rule executes an instruction a person already wrote, whereas the agent chooses which instruction applies and overrules it when the data disagrees. Executing a fixed policy is bookkeeping; authoring one in real time is the decision operators keep for themselves. Even Gartner, while forecasting far more autonomy ahead, urges that higher-stakes calls augment human judgment rather than replace it.

Strip away the branding and Wallie is an argument about accountability dressed as a forecasting tool.

The decision that matters here is organizational, not technical: whether a firm will let a person answer for a call they did not make. An agent that halts replenishment to a store is also an agent that can starve a location that would have sold through, and when it errs the loss lands on a manager who never chose it. Firms that want the margin will have to build what the vendor decks leave out: a way to hold someone answerable for a decision the machine authored, or the nerve to accept that no one is. If they can carry that, the economics are real, and Gartner’s forecast that 60 percent of supply-chain disruptions will resolve without human intervention by 2031 will describe fashion too. If they cannot, Wallie reverts to what inventory software has always been: a very confident dashboard whose advice a planner still has to approve. The agent has arrived with opinions; the one line it cannot sign is the one that says who answers for them.