Everyone’s talking about agentic AI, but why are so few actually scaling it?
In this Digital & AI Edge executive panel, Jen French, General Manager AI Acceleration at CommBank, Bhaskar Katta, General Manager at Westpac, Simon Kriss, ADAPT Advisor and CEO at Sovereign Australia AI and Peter Hind, Principal Research Analyst at ADAPT unpack the gap between experimentation and real-world deployment.
This reveals that the tipping point isn’t about better tech, but about discipline, design and organisational readiness.
Key takeaways:
- Scale is built, not bought: Successful agentic AI comes from incremental capability building, not leapfrogging to advanced use cases.
- Control determines value: Financial discipline, governance and visibility are critical to avoid cost blowouts and operational risk.
- There’s no universal playbook: The right approach depends on organisational culture, leadership alignment and how well people are brought into the journey.
The journey to scale is evolutionary, not a leap
Moving from POC to production requires a layered, iterative approach, not a single “big bang” deployment.
Many organisations stall because they try to jump straight to advanced agentic use cases.
In reality, scaling AI depends on building foundational capabilities, process intelligence, data understanding and incremental automation, before introducing more autonomous systems.
Bhaskar explains that Westpac’s journey started with RPA, evolved through document processing and knowledge tools, and only then moved into agentic AI, allowing change management, stakeholder buy-in and trust to build along the way before scaling.
The real risk isn’t adoption, it’s uncontrolled scale
Without governance and financial discipline, agentic AI can quickly become a cost and risk multiplier rather than a productivity driver.
Agentic systems operate at machine speed, amplifying both value and risk.
From token consumption blowouts to model dependencies and system fragility, organisations must actively manage visibility, cost and control.
Panellists highlighted “token shock” as a key emerging risk, where autonomous loops consume massive compute spend in minutes.
At the same time, fragmented experimentation can lead to uncontrolled sprawl, forcing organisations to balance innovation with central oversight and decision discipline.
Success depends on people, design and organisational self-awareness
The organisations that scale agentic AI successfully are those that redesign work, involve people early and align with their own cultural DNA.
There is no single “right” approach, (data-first, process-first, experimentation-first or governance-first), all work depending on the organisation.
What matters is clarity of outcomes, internal alignment and the ability to bring people on the journey.
Jen emphasises starting with outcomes and redesigning processes using design thinking, while ensuring customer service teams are involved early to build trust.
Meanwhile, leaders stress aligning strategy to organisational culture, whether risk-averse or entrepreneurial, rather than copying others’ approaches.
This piece was a paid content partnership between Commonwealth Bank of Australia and ADAPT.