# Agents expose commodity, not brand > When an AI agent compares for the shopper, retailers that win only on price and availability are exposed. A first-hand test of ASOS Stylist in ChatGPT. Source: https://barkingcat.ai/thinking/agents-expose-commodity/ ## Thinking · October 2026 · Position When an AI agent compares on the customer’s behalf, the retailers most exposed are the ones whose only reason to be chosen was availability and price. The brands customers ask for by name are defended. The agent works for the shopper, and it will say so with its links. In May 2026 ASOS launched ASOS Stylist as an app inside ChatGPT, built with Bambuser, which turns the ASOS catalogue and video library into data the assistant can return as shoppable content. It is a well-conceived integration and a reasonable bet. I tested it in its first week. I asked it to build a smart casual wardrobe for a middle-aged man. It returned a Mango blazer, Jack & Jones shirts and Thomas Crick trainers, which was competent. Then it reached fragrance, recommended Tom Ford Oud Wood, and linked me straight to tomfordbeauty.co.uk. ASOS does not stock it. Within three minutes, a stylist ASOS had placed inside ChatGPT had sent a customer to another retailer. The stylist did exactly what a good adviser should. It gave the best answer for the person asking. The best answer was somewhere else. ## The principle and its missing clause A month later, ASOS’s chief technology officer, Przemek Czarnecki, set out the single principle behind the company’s response to agentic commerce in an interview with McKinsey: agentic shoppers and human shoppers must be treated equally well. It is a good principle and ASOS is further ahead for holding it. My test shows the clause it needs next. Treat agentic and human shoppers equally well, and make sure the agent knows whose shop it is standing in. A human stylist employed by ASOS navigates the gap between the shopper’s interest and the retailer’s without being told. An agent built on someone else’s model does not, unless it is taught. ## What an agent actually compares An agent shopping on someone’s behalf reduces a decision to the attributes it can read and rank. So it matters what shoppers say they choose retailers on. In a January 2026 survey of 3,004 US consumers for McKinsey and ICSC, the top reasons were being able to reliably find what they want (37%), a convenient location (33%) and a wide range of price points (25%). Experiential elements came last, at 5%. Every one of the leading three is a column in a comparison table. Stock, distance or delivery time, price. An agent reads them faster and more completely than any shopper, and it has no habit to break. The practice’s reading, which is inference rather than anything the survey measured: a retailer whose only reason to be chosen sits in those columns is a row the agent can replace with a cheaper identical row, and the relationship it thought it had turns out to have been a habit. The shopper is already halfway there. McKinsey’s State of the Consumer 2026, surveying 4,863 consumers across five markets including the UK, found more than three quarters trading down in some form and 82% keeping things longer before replacing them. In essentials, they optimise within the category: switching brands, changing pack sizes, chasing promotions. A price-comparing agent meets a customer who was already doing the comparing. ## Where brand still holds, and where it does not The defended position is the one the customer creates before the agent starts. If the prompt names the brand, the agent is an errand-runner; it fetches the thing asked for. Considered, emotionally held categories are well protected for that reason, and the agentic threat in them is smaller than the conference circuit suggests. But the defence has a gap. McKinsey’s report also cites third-party tracking of around 2.6 million citations for 25 consumer goods brands across ChatGPT, Gemini and Perplexity in the UK and US: about 1% of the sources the models cited were the brands’ own websites (data from XEO360, a vendor in the citation-tracking business). Being asked for by name protects the brand at the query. What the model then says about it is largely written by retailers, reviewers and forums. So brand is the defence, and it is a defence the brand does not fully control once an agent is the narrator. ## The operator question Take price and availability out of your proposition and look at what is left. Is there anything a customer would ask an agent for by name? If there is, the work is making sure the product data, the reviews and the third-party descriptions an agent reads tell the same story your stores do. If there is not, the agent has simply found the problem earlier than the next downturn would have. ## Sources - ASOS plc, ASOS launches ASOS Stylist app in ChatGPT, 20 May 2026; Bambuser release of the same date (Bambuser is a technology vendor and co-announcer). - Barking Cat, first-hand test of ASOS Stylist in ChatGPT, May 2026; reported by Retail Technology Innovation Hub, 27 May 2026. - McKinsey, The rise of the agentic shopper: ASOS’s AI investment, interview with Przemek Czarnecki, 18 June 2026. - McKinsey and ICSC, Shopping in the age of AI: Redefining stores for a new era, April 2026, Exhibit 2 (ICSC Consumer Survey, 7 to 11 January 2026, n = 3,004, US). - McKinsey, State of the Consumer 2026: When tech acceleration and cost pressures collide, June 2026: trade-down and replacement behaviour (n = 4,863, Brazil, France, Germany, UK, US, March to April 2026); LLM citation sources (XEO360 data, October 2025 to May 2026, consumer goods brands only; vendor data). ## Contact steve@barkingcat.ai Barking Cat Advisory Limited. Registered in England and Wales, company number 17338003.