As they seek to build Australia's most advanced agentic AI business, Lendi's AI team is transparent about the fact that genuine transformation is messy and frought. For instance, agentifying its flagship home loan process (codenamed Guardian) was an at times bruising, highly commercial education in what agentic really is and what it definitely isn’t. Executives have outlined a number of crucial lessons they've learned, including one of the real unlocks: Agents must be trained to sell, not just to answer. Lendi shipped a “beautiful” agent with bad outcomes, because it asked for documents too early and never demonstrated value. The fix wasn’t just better prompting. Instead, it was sales logic in the funnel, specialist agents up front, and relentless measurement. These are the Guardian lessons in full: unlearn the old ways, design the agent workforce like a business, keep the plumbing deterministic, put value before friction, and treat governance as a product feature, because without it, your Mozart-like swarm turns into Trump, and you won’t know what it’s doing from one minute to the next.

Marketers pioneering agentic AI are focusing on three distinct agent classes: Agents for marketers, agents for customers, and most disruptively, agents of customers. Each reconfigures the marketer’s role, and in some cases, forces a rethink of what brand means in a world where the first audience is artificial. But it's the third domain where the real brand shift lands. Agents of customers are tools that will judge brand content before a human ever sees it. For marketers terrified by the implications, martech expert Scott Brinker has some advice: Chill, you have been doing this for years with search, you just didn't notice. But don't lag behind. This year's Martech for 2026 survey by Brinker and Frans Riemersma reveals increasing agent deployment, especially among more tech savvy organisations. As Brinker puts it, marketers used to optimise for a single agent. Now they face a council of machines, each one acting on behalf of someone else. The question isn’t whether marketers are ready. It’s whether their brands will even make the cut.

Nathan Alexander is a chief information officer who thinks like a CMO should think. He’s rewired one of Australia’s oldest firms – and it was painful. But in the process McPherson’s found 45 per cent of the $30 million it spends on trade promotions with pharmacies like Chemist Warehouse, Priceline, TerryWhite Chemmart and long tail of independents was delivering negative ROI. So the potential payback is big. Now McPherson’s is pushing into ecom to save a packet on the data it has to buy across 400-plus SKUs, while using sharper promotions data and agentic AI to prepare account managers for “tough conversations” with big chemist chains about what’s working, what’s not – and how to do it better.

A sweeping field study on one of the world’s largest cross-border ecommerce platforms – unnamed by the researchers at Columbia and Zhejiang Universities, but with hundreds of millions of buyers, hundreds of thousands of sellers, across 20 countries – has delivered the cleanest real-world evidence yet that AI lifts revenues not by being clever, but by bulldozing friction. The numbers are startling in their simplicity. Pre-sale chatbots, previously a dead channel, drove sales by 11–16% and lifted conversion more than 20%. Search translation and AI-generated descriptions added another 2–3% conversion. Stitch those early-stage workflows together and the platform banked roughly $5 in extra revenue per consumer in a single year. At this scale, that’s not incremental. And the impact wasn’t felt by the power sellers or heavy users. It was the smallest merchants and most hesitant shoppers, the people markets usually leave behind, who saw the biggest gains. In a sector chasing “autonomous commerce,” the lesson is blunt: AI isn’t winning by persuasion. It’s winning by making ecommerce simple, fast, and easy, and by proving that productivity can rise from the demand side as sharply as the supply.

If the support team's tech breaks during a Bledisloe Cup match, the game breaks too. That’s the blunt reality facing sports analytics firm Catapult, whose customers include elite teams where data latency isn’t an inconvenience,  it’s a tactical liability. When a cluster of sensors fails mid-match, there’s no time for “have you tried turning it off and on again?” The company’s response architecture has to match the velocity and the ferocity of the sport. So it is leaning into agentic AI as a toolkit for decisioning, escalation, and multilingual triage at speed. Catapult’s experience confirms agentic AI needs much more than smart generative models to succeed. According to Catapult global head of CX Alexandra Finster-Rowen, the magic only happens when you get the plumbing right. The real lifts come from disciplined inputs: Workflows, governance, content and constraints.

People no longer want to sit around eating and drinking, they want to do something while they’re at it. That shift has seen multi-venue operator NightOwl double revenues in the last 12 months via pubs and clubs where high tech darts and bowling are the big draw. Boss David Heaton aims to treble revenues over the next five years with the same lean headcount – but he couldn’t do that without first rebuilding the entire martech stack and plugging into agentic AI to deflect massive enquiry volumes – and start upselling. Painful, but now the firm is templating that experience, warts and all, as a service to other operators that will inevitably have to go through the same pain.

Sixty interviews over six months. Mi3’s Agentic AI report unpacks expert views on where the market is today, where it’s headed – and the risks and opportunities now facing marketers, IT, operations, management and boards. How quickly and how successfully agentic AI adoption matures from here hinges on orchestration – the rules of engagement – and governance. Done well, the rewards scale dramatically. Without them, agentic AI won’t amplify intelligence; it will automate disorder.