I was recently sitting with a senior seasoned retail executive, and our conversation wandered into agentic commerce and trust. It got me thinking.
There’s a lot of talk about autonomous agents making purchases on our behalf, and the repetitive, low-involvement purchases are likely to be the first delegated. But many shoppers still don’t trust an agent to complete a transaction for them. That got me thinking about the varying levels of trust in the agentic commerce space, because trust will be the deciding factor in adoption. And trust, I believe, is a continuum, and the issue is layered.
A recent Commerce, PayPal and Logica study of 1000 Australians reinforced just how far we have to go when it comes to autonomous agents making decisions without human intervention, with just 4% of Australians wanting help to complete the checkout process today as part of an agentic shopping experience. That’s likely to change as trust builds, but it’s a stark reminder of ambitions vs today’s reality.
The multi-layered trust gap
Trust in agentic commerce comes in various forms. There are macro issues of security and data, and then there are more nuanced issues around trust related to who is completing the transaction on behalf of the shopper. My firm belief is that the retailer is at a natural disadvantage here, because it lacks independence in the process.
Security: The Commerce, PayPal & Logica study from June found Australians are notably more concerned about security than shoppers in the US and UK, with unauthorised purchases, bank account breaches and personal data security ranking among the top barriers of adoption. If I don’t trust the protections are in place, I’m not delegating.
Institutional trust: How I feel about a sector and the providers within it is another dimension that impacts trust. Major supermarkets, for instance, have faced real scrutiny over alleged price gouging. So even though the weekly shop is mundane and repetitive, exactly the kind of task agentic commerce should win first, would shoppers trust the majors to buy for them, or lean toward an independent platform that compares across retailers? Any retailer entering the agentic era brings its existing reputation with it. Brands that have built strong consumer trust will have an advantage, while those facing credibility challenges may find shoppers less willing to delegate decisions.
Corporate self-interest: The third layer is about who the agent is actually optimising for. Even if I trust the sector and the security is airtight, I still need to believe the agent is acting in my interest rather than someone else’s. A retailer’s incentive is visible and direct, it profits from steering me toward its own products, higher-margin lines or stock it needs to move. Consumer AI platforms don’t have that same relationship to any one product, but they aren’t neutral either. We’re yet to see how sponsored placement and other media arrangements shape what an assistant recommends over time and as it is less visible, it is less likely to be front of mind for shoppers.
Guarantees and protections will reduce inertia and drive adoption. For Australians, that means control in the interim: a final say before the transaction is made, and the confidence that if something goes wrong, there’s recourse.
✨ Does Joy give us joy?
Kmart launched an AI shopping assistant. So naturally, I had to put it to the test.
Joy is Kmart’s new conversational shopping assistant, built on Google Cloud’s Gemini Enterprise for Customer Experience and rolled out inside the Kmart app in June. It can take a photo or a text prompt and turn it into product recommendations, layer on a virtual try on for clothing, and preview furniture in your own space through a “see it in my space” function. Kmart’s Chief Customer Officer Bernard Wilson has described it as a way to help customers narrow their choices and find what is right for them, “from discovery to decision making, wherever they are.”
That is a big claim for a retail chatbot. So I gave Joy three everyday briefs and watched what it did with them.
Why I tested it
Every retailer seems to be racing toward the same idea right now. Woolworths has Olive, Bunnings has Buddy, Amazon has just folded Rufus into Alexa for Shopping, and ChatGPT and Google are both fighting for the moment a shopper starts typing “gifts for” into a search bar. The pitch is always the same: less scrolling, more finding. I wanted to know whether Joy actually delivers that, or whether it is another AI layer bolted on top of the same catalogue.
What I asked
I ran Joy through three real shopping scenarios: a birthday party for a four year old, redecorating a bedroom, and kitting out a first camping trip for two.
What it got right
Camping was Joy’s best moment. I told it I was going camping this weekend with no gear at all, and it built out a genuine plan in phases, tents and sleeping bags first, then cooking equipment. When I added that there were two of us and we wanted to cook burgers, it adjusted the whole list to two person tents and double sleeping options, then pulled together a burger station complete with a portable grill, a burger press and BBQ tools. It even offered to follow up with plates, cutlery or lighting for an evening meal by the fire. This is where a conversational assistant should shine: a big, slightly overwhelming task broken into a sequence a first timer can actually follow, with the product recommendations doing real work at each step.
What it struggled with
The other two briefs showed the cracks. When I said I was shopping for a four year old’s party, Joy gave me decorations, balloons and bunting. When I followed up that it was for a girl who likes mermaids, it happily pivoted to mermaid dolls, hair clips and blankets, but it had dropped the party context entirely and never circled back to the birthday theme, and it had also lost track of her age, asking me afterwards what age range I was shopping for. Context that I had given it two prompts earlier had simply fallen out of the conversation.
The bedroom redecoration followed a similar pattern. Joy asked about my style preference, I said neutral and modern, and it responded with a three phase plan (foundation and storage, comfort and textiles, finishing touches) that was close to a word for word repeat of the generic plan it had already given me before I mentioned my preferences at all. It then offered to generate an image showing how the pieces might look together in a bedroom setting. The image itself was pleasant enough, a soft, neutral, styled room, but it was not clear what it added over an existing product shot or a simple room mockup, and the disclaimer underneath reminded me the appearance and size of products in the image may not match the real thing anyway.
The other shortfall is that Joy’s recommendations take you through to the product page rather than adding straight to cart, so it is guiding discovery rather than completing the transaction for you.
Would I actually use it?
For a task with a clear, linear checklist, like camping, yes, without much hesitation. For anything that depends on the assistant holding two or three details in its memory at once, like a themed kids’ party, not yet. The core promise of a conversational assistant is that you should not have to repeat yourself, and right now Joy asks you to.
The verdict
Scoring Joy across the fundamentals, it lands at 6.8 out of 10. It understands intent (4/5) and is genuinely strong on product recommendations and shopping inspiration (4/5 each), pulling in relevant, well priced options at every turn. It is fair on solving real customer problems (3/5), useful for a straightforward brief but not yet for a layered one. Context and memory is where it falls short at just 2/5, which is where it let it down the most across both the party and the bedroom tests.
What this means for retailers
Compare that to where the category is actually heading. Amazon rebranded Rufus to Alexa for Shopping in May, explicitly to fuse Rufus’s product knowledge with Alexa’s memory of a customer’s stated preferences, so a parent who has already mentioned their kids’ ages and interests gets that folded into every future recommendation without repeating it. ChatGPT, meanwhile, tried the opposite bet, in app checkout through Instant Checkout, and pulled it back in March after fewer than fifteen Shopify merchants ever went live, pivoting instead to richer visual discovery and comparison tools. Google has taken a similar discovery-first approach through its own commerce protocol with Walmart, Target and Shopify on board. The pattern across all three is that the winners are not necessarily the ones with the flashiest generative feature. They are the ones that remember what you told them and get the fundamentals of search and comparison right.
That is the test Joy still has to pass. Decorations and camping gear are a reasonably forgiving category to launch in, low stakes, low complexity, easy wins. But a shopping assistant only earns its keep once it can hold a customer’s context across a longer, messier conversation, the kind real shopping trips actually are.
And that points to the bigger structural question every retailer chasing this trend needs to answer. Having a clever assistant that only lives on your own site or app is not, on its own, an AI strategy. Shoppers are increasingly starting their research in ChatGPT or Gemini before they ever open a retailer’s app, so the real competitive question is not just “how good is our assistant” but “how discoverable and well represented are we inside everyone else’s.”
Joy is a promising first step and clearly built on solid infrastructure. But right now, it gives more joy when the job is simple than when it is personal.
💬 So what’s been happening in the digital, data & eComm space?
📊 Temu tops the charts while virtual assistants reshape product discovery. Despite all the attention on AI, social commerce and marketplaces, the IAB Australia’s latest Commerce & Discovery Report shows retailer websites remain central to the shopping journey, with 22.1 million Australians visiting an online retail or commerce brand in May alone. Temu topped the rankings ahead of Amazon and Woolworths, while supermarket websites recorded a 35% increase in time spent year-on-year. At the same time, the way consumers research products is becoming increasingly conversational, with 54% of Australians using AI assistants to discover products and 45% using live chat for research & discovery (use of live chat rises to 64% among 18–39 year olds). The message is clear: retailer websites remain critical, but the path to purchase and engagement approach is becoming increasingly conversational.
♻️ Australia’s resale market is entering a new competitive era. In just a few months, eBay removed transaction fees for eligible private sellers, Vinted launched locally, and Depop announced it will remove seller fees for Australian sellers from 22 July. Together, these moves highlight an important marketplace dynamic: inventory is the competitive advantage. The platform with the most attractive range of products is typically the platform that attracts the most buyers, making sellers the first battleground in the fight for market share. Fee reductions are therefore less about generosity and more about accelerating marketplace growth. The opportunity is significant, with Millennials and Gen Z increasingly embracing resale as a normal part of how they shop - driven by value, individuality and circular consumption rather than sustainability alone. For retailers and brands, it’s another reminder that recommerce is no longer a niche trend. It’s becoming an increasingly important part of the retail landscape, reshaping how consumers think about value, extending product lifecycles and creating new competition for the sale of new products.
🤖 The rapid growth of AI is reflected in the sheer volume of research now being published on the category. In recent weeks alone, we’ve seen reports analysing everything from AI market share and app adoption through to shopping behaviour, referral traffic and consumer trust. Sensor Tower’s State of AI 2026 report reinforces just how quickly AI has moved from experimentation to everyday behaviour. ChatGPT reached one billion downloads faster than any app in history, while global time spent using generative AI apps is projected to more than double year-on-year, reaching 36 billion hours in the first half of 2026. Consumer spending on AI apps is also expected to exceed US$4 billion over the same period, highlighting that AI is no longer just attracting curiosity, it’s becoming a utility people are willing to pay for. The next phase of AI won’t be defined by whether consumers adopt it, but by which brands and retailers become part of the experiences consumers are already using every day.
🦊 So who am I?
I’m an advisor, trainer and thought leader with 25+ years’ experience across digital, marketing, loyalty and commerce, working with consumer, FMCG and retail brands. I’m also the host of Unpacking the Digital Shelf – APAC Edition and have been recognised as a RETHINK Retail Top Retail Expert (2026), and Top 20 CMO (2018).
Before you go…
🧭 Need strategic clarity or expert advice? Let’s chat.
🛠️ Looking to build capability in your team? Explore training & workshops.
🎤 Planning an event or offsite? Book a keynote or thought leadership session.
Get in touch at teresa@arkticfox.io or find out more here, and don’t forget to follow along on LinkedIn and Substack.
Trusting an AI to buy on your behalf is fundamentally a question of liability. You are entirely right that distrust in agentic commerce will not be equal across the board. Reputation and corporate self-interest dictate the entire market dynamic. I agree completely with this premise in my article "Everyone Is Shopping for Agentic AI Use Cases. Nobody Is Governing the Authority to Act." We are rushing to give software the autonomy to execute transactions before establishing the security frameworks required to govern those decisions. Unchecked delegation is a breach waiting to happen.
https://cyrilsimonnet.substack.com/p/everyone-is-shopping-for-agentic