The biggest competitive shift in Australian mortgage broking this decade is not coming from a new bank, a fintech challenger, or a rate war. It is coming from inside the channel itself.

Lendi Group, the owner of Aussie and one of the country’s largest broking businesses, has publicly committed to an “AI-native” future. The ambition is striking: by June 2026, the group wants every mortgage to be processed, monitored and managed by agentic AI, with humans stepping in only for the complex, relationship-heavy work that machines cannot do well. The internal program reportedly carries the codename Project Aurora, and Aussie’s mortgage process is set to become fully AI-led. Lendi’s CEO has framed the model not as machines replacing people, but as “agents managing humans” – software agents acting as coaches and orchestrators rather than bosses.

Whether or not that exact timeline holds, the direction is now set, and it is worth taking seriously. When an operator of that scale rewires its entire process around autonomous AI agents, it changes the competitive baseline for every independent broker in the country – the solo operator working from a home office and the boutique principal with three writers and a support team alike.

This is not a panic piece. It is a strategy piece. The honest answer is that a fully-agentic aggregator will, in time, beat most humans on speed and cost for vanilla loans. But that is not where the independent broker’s value has ever truly lived. The task now is to understand exactly what an agentic model can and cannot do – and to deliberately build a moat around what it cannot.

What “fully-agentic” actually means

It helps to be precise. A chatbot answers questions. An automation runs a fixed rule. An agent is different again: it takes a goal, breaks it into steps, uses tools, makes decisions within guardrails, and chases an outcome with limited human input.

In a mortgage context, a mature agentic stack could plausibly do a great deal of the file – collect and verify documents, read payslips and statements, pre-assess serviceability against lender policy, shortlist products, draft the application, chase conditions, monitor the loan post-settlement for repricing triggers, and prompt the customer at the right moments. The Lendi vision is that agents run this end-to-end pipeline and escalate to a human only when the file is genuinely non-standard.

You will not out-process a machine built specifically to process. The winning move is to be clear-eyed about where the machine is weak.

Where the agentic model is genuinely weak

Three weaknesses are structural, not temporary – and they are exactly where broker value concentrates.

1. Complex and non-vanilla scenarios

Self-employed income with add-backs and a messy trust structure. A construction loan with progress payments. A guarantor arrangement inside a blended family. A client recovering from a credit event who needs a near-prime lender and a documented exit plan. These files are not edge cases for many brokers – they are the bread and butter. They demand judgment, lender-by-lender policy nuance, negotiation, and the confidence to structure something a policy engine would simply decline. The human a system escalates to may as well be you.

2. Genuine relationships and trust

A mortgage is often the largest financial decision a household ever makes, made under stress. People do not refer their sister to an algorithm. They refer her to the broker who sat at their kitchen table, explained offset versus redraw in plain English, and answered the phone on a Sunday when the finance clause was due to expire. Trust is earned in moments a pipeline cannot schedule.

3. Accountability and documented advice

This is the most underrated moat of all, and it is uniquely Australian. Under the Best Interest Duty, a broker must act in the client’s best interests and evidence why a recommendation was genuinely suitable. BID-grade advice is not a product match; it is a documented chain of reasoning – needs identified, options considered, trade-offs explained, rationale recorded. A named, credentialed human standing behind that advice, accountable to ASIC and to the client, is something a fully automated funnel cannot easily offer. The duty does not disappear because an agent did the legwork; someone still has to own the recommendation.

The four pillars of the independent moat

A practical framework: keep, augment, automate

Defending the moat does not mean rejecting AI. The broker who refuses these tools will lose to the one who uses them well. Let AI do what it does best so your scarce hours go to the four pillars. Sort every task in your process:

The test is simple. If automating a task frees you to do more relationship and complexity work, automate it. If a task is where trust is built or judgment is exercised, keep it human and resource it properly. Most lost clients trace back to getting that sort wrong.

What to do in the next twelve months

Audit your book. Identify your most complex, highest-trust, most referral-active clients and double down on them. Then identify your most commoditised, transactional volume and decide honestly whether it is worth defending or worth automating.

Upgrade your documentation. If your BID file notes are thin, fix that before it becomes a liability and while it can still be a differentiator. A consistent, advice-led record is both your compliance armour and your story to clients about why a human matters.

Adopt deliberately. Bring AI into your own workflow on your terms, starting with commodity tasks. The goal is leverage – more capacity for the human work that wins, not a cheaper version of the same race the aggregators are built to win.

The bottom line

A fully-agentic aggregator is a formidable competitor for simple loans, and pretending otherwise helps no one. But broking has never really been a processing business; it has been a trust business that happens to involve a lot of processing. As the processing is automated away, the trust becomes more valuable, not less. The brokers who thrive will let the machine take the commodity, and stake their name on the four things it cannot replicate: complexity, relationship, advice and network.