Macquarie brings agentic SRE to its digital bank

Macquarie’s banking division is taking reliability cues from Google, adopting its site reliability engineering principles and using AI agents to enforce them.

Macquarie Group's Phillip Grasso-Nguyen.

The effort is being spearheaded by divisional director and head of engineering excellence and reliability Phillip Grasso-Nguyen, a former Googler executive and CBA engineering lead.

Speaking at Dynatrace Perform 2026 at the end of last month, Grasso-Nguyen said that Macquarie had demonstrated that a bank can achieve “tech company-level reliability”.

“We exceeded our targets and hit 99.98 percent availability and reliability for our customers, with a faster detection rate at 79 percent, with 59 percent fewer critical incidents,” he said.

Grasso-Nguyen said Macquarie set a goal to “become the best tech company in financial services.”

“I thought this would be easy coming from Google, but I was wrong,” he said.

“Tech companies treat their systems like cattle. They’re uniform and they’re replaceable, done at scale. Enterprises and banks treat their systems like pets, with a lot of care and support, a lot of complexity and a lot of personality.

“I had to shift my mindset from herding cattle to looking after pets. I had to embrace the complexity explosion that exists in enterprises; and I had to focus on one thing that mattered the most, which is our customers.”

Underpinning Macquarie’s retail digital bank is a cloud- and Kubernetes-based stack comprising “over 6000 services”.

Grasso-Nguyen said the bank typically deploys in excess of “8000-plus changes and releases a year”, with the work completed by 1500 engineers.

“We have no appetite for customer disruption. So what does this mean? We’re not allowed to have scheduled outages, ever.

“So can you imagine how the core banking team felt and how much fun they had trying to replace parts of their core banking system last year,” Grasso-Nguyen said.

The bank ultimately needed to be both “reliable and innovative” in its approach to engineering, and this ushered in a focus on site reliability engineering (SRE).

Grasso-Nguyen said that the typical approach to having a reliability incident “is to slow down” and add process.

“So, often that’s done through CABs [change advisory boards],” he said. 

A change advisory board is a decision-making body that oversees change management in organisations.

With a background at Google, however, Grasso-Nguyen sought to take a different approach.

“We applied the Google SRE methodologies and principles, so that meant no CABs, no change boards, [and] smaller changes more often, with a lower blast radius,” he said.

“By adapting engineering ways of working from the tech companies, we managed to get 90 percent on-time project delivery, with 80 percent less defects. 

“We proved that you don’t have to trade off reliability for innovation.”

Macquarie supports its reliability, change and incident management using an application performance management (APM) and observability stack supplied by Dynatrace.

This has helped the bank meet its reliability objectives. In Grasso-Nguyen’s words, Macquarie is "not just looking for problems but making sure we’re avoiding them getting into production.”

Macquarie is using SRE AI agents – both Dynatrace’s and its own – to assist with all aspects of the capability.

“We have AI agents that are doing automatic diagnostics [and] running runbooks. We’re using the Dynatrace API to power our AI agents,” Grasso-Nguyen said.

“[For example], we have an automated release verification agent that acts as an SRE gatekeeper, automatically catching problems and change failures. 

“Now, in every incident, we use AI agents.

“Where previously we had a large number of teams in the triage, now we have operators with assisted AI agents that can get to problems in minutes. We only call in the relevant teams now so it means that we spend a lot less time diagnosing and repairing, and we don’t waste as much time as we used to.”

While Macquarie has built its own agents where out-of-the-box capabilities do not exist, this is likely to change.

At the conference where Grasso-Nguyen spoke, Dynatrace unveiled an agentic operations system called Dynatrace Intelligence.

Grasso-Nguyen said Macquarie is now “hoping to partner” more closely with the vendor “to evolve our capability.”

“If we don’t need to build it, why build it ourselves?” he said.

Customer-facing agents

Grasso-Nguyen said that Dynatrace is also helping the bank to deliver customer-facing AI experiences.

Macquarie launched a customer-facing AI agent called Q in mid-January.

Grasso-Nguyen said that Dynatrace is monitoring “usage, effectiveness, cost and downstream dependency health” of Q.

“Observability in our AI workloads is critical,” Grasso-Nguyen said.

“Having that [observability in place] gave us the confidence to release [Q].”