August 5, 2026
“Agent” has quietly become the most overused word in retail technology. It’s thrown around in every meeting and every vendor deck, and rarely defined. But the differences between one kind of agent and another are not academic. They decide whose side the AI is actually on when it’s standing between you and your customer.
The advocate is an agent in the commercial sense – like a travel agent acting for you. It’s permissioned and declared: it works on your explicit authority and shares your values. The spy is an agent in the intelligence sense: a covert operator whose real allegiance is hidden, and whose interests can quietly conflict with the customer’s. The actor is an agent in the fullest sense of the word: an agent capable of independent judgement based on an inferred model of your values, rather than explicit instructions. It’s the most autonomous, and for now, the most theoretical.
Most agents a retailer builds for itself are advocates: clear allegiance, bounded scope. The commercially ambiguous ones are built by the platforms that sit between retailers and customers, such as Google’s shopping agents, ChatGPT in discovery mode, Amazon’s Alexa for Shopping. They claim to represent the customer, but the platforms behind them have investors who expect returns and advertisers who pay for placement – paying the platform, not the retailer whose products the agent is selecting between.
That’s the double agent. The customer believes they have an advocate, but they really have a spy. The architecture looks legitimate, and it really is acting on the customer’s behalf – it’s just also acting on somebody else’s. When Amazon’s Rufus shops on a customer’s behalf through Scheduled Actions, the customer believes it’s working for them. But it’s operated by Amazon, whose interests aren’t identical to theirs. The conflict isn’t a side effect; it’s part of the design.
From the book: The Agent Taxonomy – who does your agent serve? (The Joshua Problem, Ch.1)
All agent types come down to one test you can apply to any agent touching your business: whose interests does it serve, and who authorised it? An advocate deserves your data; a spy doesn’t. For platform agents specifically, the book puts it like this:
“Before granting any platform agent access to our systems and our customers, can we answer the three questions posed earlier in the book: what data does it access, how does it rank our products, and who is accountable when something goes wrong? If we cannot answer all three, we are extending trust we have not verified.”
“Agent” isn’t one thing: it’s at least three (and possibly four), and they don’t all answer to the same master. The retailers who treat that distinction as a contractual question, not a technical one, are the ones who’ll avoid handing their customers, and their data, to an agent quietly working for someone else.
This is one thread of a much bigger argument.
Paul is a Retail Technology Strategy Consultant, who recently joined Equal Experts after spending the last 20+ years working for retailers across the globe, including positions as CTO at Halfords and Chief Architect at New Look, Primark, Marks & Spencer, and Argos in Shanghai. Connect with Paul on LinkedIn.
All business models have their pros and cons. But, when you consider the type of problems we help our clients to solve at Equal Experts, it’s worth thinking about the level of experience and the best consultancy approach to solve them.
If you’d like to find out more about working with us – get in touch. We’d love to hear from you.
