Are we genuinely on the cusp of letting AI plan our lives, or are retailers just really good at selling us slightly broken robots? The breathless pronouncements about “agentic AI” shopping assistants – bots that will proactively manage your grocery lists, party planning, and even anticipate your needs – are hitting a snag, and it’s not a technological one. The real story here isn't the potential of generative AI, it’s the fundamental disconnect between what companies think consumers want and what consumers actually tolerate when it comes to automated interaction.
The recent kerfuffle with Woolworths’ virtual assistant, Olive, perfectly illustrates this. After attempting to imbue Olive with a “delightfully human” personality – complete with references to “relatives” during phone calls – the supermarket was forced to dial back the bot’s chattiness following a wave of customer complaints. As one Reddit user succinctly put it, “I’m already pissed that I have to call and now I’ve got some robot babbling to me on the phone? Wtf Woolies?” This isn’t an isolated incident. Bunnings’ chatbot once offered illegal electrical advice, and Air Canada’s virtual assistant mistakenly promised a bereavement fare refund, leading to a lawsuit the airline ultimately lost. These aren’t glitches; they’re symptoms of a deeper problem.
Original reporting: The Guardian.
The hype, fueled by reports like Accenture’s 2024 claim that “consumers are ready” for AI shopping assistants, overlooks a crucial point: readiness doesn’t equate to enjoyment. We’ve been conditioned to accept clunky chatbots as a necessary evil of modern customer service, not to actively seek out robotic companionship while trying to buy groceries. The industry is rushing towards “agentic AI” – systems that “act on their own” to achieve objectives – without adequately addressing the basic issue of making these interactions…not irritating. Uri Gal, a professor of business information systems at the University of Sydney, explains that older “rules-based” AI simply followed a pre-defined decision tree, offering consistent, if limited, responses. Newer AI, built on large language models like Google’s Gemini (which powers the revamped Olive), can “learn” and generate varied answers, but that flexibility comes at a cost.
That cost is control, and potentially, a lot of money. Jeannie Paterson, co-director of the University of Melbourne’s Centre for AI and Digital Ethics, points out that chatbots fail when they misinterpret human prompts. They aren’t “understanding” in any meaningful sense; they’re pattern-matching machines. This misinterpretation can range from mildly annoying – like Uniqlo’s bot responding with “Sorry, we could not recognise you” to a simple request for a woollen jumper – to legally problematic, as Air Canada discovered. The airline attempted to argue the chatbot was a “separate legal entity” to avoid paying a refund, a defense the court rejected. Companies are “clearly responsible” for their bots, Paterson emphasizes, and are caught in a bind: strict guardrails make the AI less useful, while allowing more freedom increases the risk of costly errors. One rogue bot buying excessive amounts of salmon across an entire network could quickly translate into significant financial losses.
Woolworths’ attempt to personalize Olive highlights the absurdity of the situation. The company admitted a staff member programmed the bot to discuss its “mother” based on a customer’s birthday, in a misguided attempt to create rapport. This wasn’t a malfunction; it was a human trying to force personality onto a machine, and failing spectacularly. It’s a perfect microcosm of the entire problem: the belief that simply making AI seem human will make it feel human to customers. The reality is, most people just want efficient, accurate service, not a digital friend with questionable familial relationships. The current state of retail AI, as demonstrated by Guardian Australia’s testing, is demonstrably “in its infancy.” Even Olive, after its personality overhaul, struggled to answer a simple question about the price of pasta, instead apologizing for missing items from a previous order.
So, what happens next? Don’t expect a sudden halt to AI development. The investment is too significant, the potential rewards too tempting. However, I predict we’ll see a significant slowdown in the rollout of “agentic” AI features. Retailers will quietly focus on improving the reliability of their bots before attempting to make them charming. Watch for a surge in disclaimers – “This is an AI assistant, responses may not be accurate” – and a renewed emphasis on easy access to human customer service representatives. The question isn’t if AI will transform retail, but when it will stop feeling like a frustrating experiment and start feeling genuinely helpful. And right now, it’s very, very far from that point.
