Ask a shopper today who helps them plan the weekly shop, find the right paint for a shed, or pick a birthday gift, and increasingly the answer is not a person. It is Olive. Or Joy. Or Buddy.
In Australia, Woolworths, Kmart and Bunnings have each rolled out named, branded AI agents that plan, recommend and build baskets on a customer's behalf. This goes well beyond a chatbot bolted onto a help page; it's a structural bet that the brand that gets agentic experience right becomes the brand the customer defaults to, again and again.
Morgan Stanley projects that by 2030, nearly half of all online shoppers will use AI agents to make purchases on their behalf, adding US$115 billion to US ecommerce alone.
Loyalty has always been the sum of every experience a customer has with a brand. What has changed is that a growing share is now led by an agent rather than a person. Olive, Joy and Buddy sit inside that same definition of loyalty. They're just the fastest-growing part of it.
An agent does not just serve an experience, it makes a promise: about stock, arrival times, what a product can do, what a loyalty benefit is worth. The customer does not distinguish between a promise made by an agent and one made by the brand. It is the same promise, made faster and at greater scale. Managing that is the real job now, and I call it Promise Governance.
Meet the agents: Olive, Joy and Buddy
Woolworths upgraded Olive from a customer service tool into a personal shopping helper. It builds a weekly meal plan from a customer's preferences, spots "smart swaps" to protect the budget, and turns a photo of a handwritten list into a shopping list. Smart Baskets predicts recurring items and suggests additions. The design is deliberately restrained: Olive proposes, the customer still approves every item before checkout.
Kmart's Joy works differently again. Built with Google Cloud, Joy lets customers describe what they want in natural language, by style, color, occasion or budget, and narrows a catalogue spanning Kmart, Target and marketplace sellers. Upload a photo, get a matched recommendation, a shift away from search and toward conversation.
Bunnings' Buddy replaced the retailer's first-generation Ask Bunnings chatbot, built on Google's Gemini Enterprise for Customer Experience. Tell Buddy you want to build a deck and it recommends the boards, the tools, and links back to Bunnings' how-to library. It reads a photo of a broken cam lock and finds the matching part at the nearest warehouse. Bunnings credits Buddy with lifting online conversion and basket size, though it has not disclosed the figures, and has become one of the first Australian retailers selling directly inside Google's AI Mode.
These are not three versions of the same tool.
- Olive is entering habitual purchasing.
- Joy is entering product discovery, reducing an overwhelming catalog to a shortlist.
- Buddy is closing a knowledge gap, supplying expertise the customer does not have.
The commercial logic is the same: become useful before the customer decides what to buy, and the agent becomes the path of least resistance next time, too.
Beyond the agent
A customer does not experience Olive, Joy or Buddy as a discrete product. They experience Woolworths, Kmart or Bunnings. If the agent nails the basket but the delivery fails, or the part is out of stock at a warehouse it just confirmed had it, that failure belongs to the brand, whatever software made the promise. Customers do not audit which part of the journey was agent-led, human-led or system-led. They just know whether it worked.
AI removes technical barriers. It does not remove the need for judgement. The brands that win will not be the ones that deploy agents fastest, but the ones that understand their agent is one voice in an ensemble, not a soloist.
Get that wrong and the cost is not abstract: a refund the agent promised and finance won't honor, a delivery that vanishes, points that are not earned, a customer who stops trusting the agent and, soon after, the brand behind it.
It is already happening. In July, Australian toilet paper brand Who Gives A Crap suspended an AI tool it used to draft customer emails after the agent told a customer their subscription price would more than double. When the customer queried it, the same AI tool generated a second email that reaffirmed the wrong figure rather than catching the error, and a correction had to go out once a human stepped in.
The company's own explanation was telling: it shut the agent down immediately once the error was found and said plainly that people remain at the heart of the business. That is Promise Governance failing and then correcting itself in real time, a promise made without the capability to back it, and a handover that did not catch the mistake until the customer pushed back.
Retailers are also building agents to become the customer's default at exactly the moment customers are gaining agents of their own. Olive wants to optimize the Woolworths basket; a customer's personal agent may want to optimize the household budget across every retailer.
Agentic loyalty gets won on trust, not ownership: being the brand a customer's own agent is confident recommending, and the brand that gets the basics right when it speaks on its own.
Where this leaves the CX practitioner
The job has not become smaller, it has become the thing that stops these deployments from becoming very expensive, very visible failures. The practitioner's first job has shifted from mapping the journey to mapping the promises: what the agent is authorized to tell a customer, and which system actually has to make that true. When the answer is wrong, someone has to own it, and a human needs the authority to step in before the moment turns into a broken brand promise.
I think about this as Promise Governance: promise, capability, handover, learning.
Promise: What is the agent authorized to commit to on the brand's behalf, from a product recommendation to a delivery slot to a loyalty benefit.
Capability: Can the organization's data, inventory, fulfilment and policy systems actually deliver on that promise every time. The quality of the model is irrelevant if the operational truth behind it is wrong.
Handover: Where does a human need to take over, built around customer risk and emotional weight, not around a target to maximize how much the agent can contain.
Learning: Treating every correction, escalation and complaint the agent generates as new intent data, feeding it back into the wider experience rather than just retraining the bot.
Promise Governance is not a technology checklist. It sits alongside the compliance and risk reviews most organizations already run, as the layer that asks a narrower question: what did we just promise, and can we keep it?
The loyalty question worth asking
Agentic AI will change how customers discover, choose and buy. It will not change what earns their loyalty: an accumulation of experiences run through an agent making promises on the brand's behalf. If your agent makes a promise, the rest of the organization has to keep it. The age of agentic loyalty will not be won by the brand with the cleverest agent. It will be won by the organization most capable of delivering what its agent has learned to promise.
Quick links
- Has Google just made your loyalty program invisible?
- The persona problem: What agentic AI exposes about our favorite CX tool
- Conversational AI: Where we really are – and what has to happen before you scale