Across Australia and New Zealand, forward-thinking organisations are preparing and deploying highly impactful agentic solutions. From an architectural standpoint, the summit offered a clear look at how Google Cloud’s underlying framework is maturing to host these complex ecosystems reliably. While many of the specific consumer-facing tools are debuting in Australia first, the deployment patterns being established provide an excellent architectural roadmap for New Zealand companies looking to follow suit.
The keynote: scale of adoption and platform differentiators
The opening keynote set a strong foundation for the summit, highlighting not just the scale of the underlying compute, but the incredible speed at which organisations are adopting AI.
Monthly token processing across Google’s global infrastructure has surged from 9.7 trillion tokens in May 2024 to more than 3.2 quadrillion tokens in May 2026. That’s an astonishing 1.2 billion+ tokens per second, driven by a massive market shift toward multi-step, autonomous operations.
Rather than looking at AI as a bolt-on feature, the keynote summarised Google's true platform differentiators:
Vertically integrated stack: Native control from custom TPU hardware up to the Gemini frontier models to optimise for cost, security, and performance.
Built-in governance & security: Enterprise-grade guardrails and compliance engineered directly into the platform from day one.
Practical innovation: re-architecting customer experience
Some of the most compelling sessions focused on Gemini Enterprise for Customer Experience (GECX). Think of GECX as the spiritual successor to DialogFlow, built specifically for the agentic era. It shifts the framework away from deterministic, keyword-matching chatbots toward fluid, context-aware digital assistants.
Bunnings Australia builds "Buddy"
Bunnings Australia shared insights into their platform-level evolution, highlighted by "Buddy", their custom agentic shopping assistant built on GECX. Designed to remove friction from digital DIY journeys by helping customers identify products, check stock, and source project advice, Buddy has delivered massive business value.
The Bunnings CIO shared that since launching in April, Buddy has more than doubled online conversion rates and driven a highly measurable increase in average basket size. These agentic features are currently live in Australia, with a New Zealand launch anticipated later this year.
Woolworths Australia scales "Olive"
Woolworths Australia provided a masterclass in handling massive digital volume, supporting an app ecosystem that sees an average of more than 13 million weekly visits. Replatforming to the new GECX Agent Studio, their digital assistant "Olive" has transitioned from a legacy bot into an end-to-end shopping concierge. Olive can interpret vague meal queries, check stock levels, and instantly compile a ready-to-buy digital cart.
We were also treated to a demonstration of future multimodal capability, such as taking a photo of your fridge and asking Olive to build a weekly meal plan around those exact items, while automatically adding only the missing ingredients to your shopping basket.
Following extensive testing by Woolworths' 200,000 staff, Olive went live for customers in Australia on July 1st. To maintain customer trust, the architecture incorporates multiple levels of guardrails and "agentic judges" to eliminate hallucinations. Like Buddy, these advanced capabilities are paving the way in Australia before expanding into the NZ market.
The shift in work: evolving roles and the SDLC
Beyond consumer-facing applications, the summit provided a very grounding perspective on how agentic AI is rewriting internal operations and the software development lifecycle (SDLC). This is something our team has been contemplating and discussing, so it was a timely session to attend.
Next-Gen Dev: a disciplined life cycle
Google highlighted the risks of unstructured "vibe coding", casually prompting an AI assistant without proper design controls. While AI tools drastically compress build times, enterprise reliability requires structured engineering. By codifying requirements, architecture, and deployment plans before generating code, teams avoid massive technical debt. It’s not about abandoning traditional software discipline; it’s about automating it.
If this interests you, the following whitepaper on the “new” SDLC is worth a read: https://www.kaggle.com/whitepaper-the-new-SDLC-with-vibe-coding
Bendigo Bank: shifting from "doing" to "training"
Bendigo Bank outlined a highly pragmatic blueprint for the agent-first enterprise, moving employees away from manual spreadsheet administration (from human doing) and into roles focused on orchestrating and validating intelligent systems (to human training).
Applying this framework to an invoice validation workflow, the bank demonstrated a 4x capacity lift within the first month alongside a noticeable increase in workforce satisfaction. Their core lesson was timeless:
“The hardest problems to solve are ultimately organisational, not technical”.
Success depends on how effectively a business captures its unique context and surfaces it securely to agentic teams. This starts by democratising AI, accelerating learning, and making AI an enterprise objective owned from the board down.
Performance & protection: high reasoning and secure foundations
Heidi Health: the value of broad frontier reasoning
Digital health startup Heidi Health shared a great reminder that meaningful innovation starts by "getting angry" at a broken status quo. An interesting note from their session was the value of general-purpose frontier models over hyper-specialised models. Citing a recent Nature Medicine study, they noted that broad models like Gemini now actually outperform specialised clinical models on medical benchmarks. The deep, general reasoning that comes with frontier models handles complex, unexpected edge cases far more effectively.
A practical security baseline
To make autonomous workloads viable for the enterprise, Google highlighted a practical Minimum Security Baseline rooted in the principles of their Secure AI Framework (SAIF). Teams were urged to focus on a secure-by-default cloud foundation using three core areas of product:
Network Isolation (VPC Service Controls): Establishing strict virtual perimeters around sensitive AI projects to block unauthorised data movement and exfiltration.
Least-Privilege Identities (IAM): Implementing tight access controls for AI to ensure autonomous agents and user accounts only have the exact permissions required.
Governance & Threat Protection: Deploying tools like Model Armor (an inline "AI Firewall" to block PII data leakage and prompt injections) alongside the Agent Registry to manage the lifecycle of models and prompt playbooks.
Final thoughts: building with intention
If the Sydney Summit proved anything, it is that the enthusiasm surrounding this space is completely justified. The conversation has officially shifted from technical potential to hardened, enterprise-wide execution. The platform foundation provided by Google Cloud is mature, secure, and ready for scale.
As we help organisations navigate this space, the path to real value remains rooted in the alignment of people, process, and technology. The tools are ready; the next step for leaders across Australia and New Zealand is building with intention, solid governance, and a sharp focus on real-world outcomes.
