Executive summary
Published on Moxie Insights, this piece features Mantel’s Head of AI Solutions Vihan Patel on how leading Australian enterprises are moving from agentic AI experimentation to scaled deployment. Vihan outlines Mantel’s four strategic priorities: self-healing agents, unit economics, production observability, and two-speed adoption, and draws on live client work spanning HR automation, financial reconciliations, frontline operations, and agentic commerce.
His sharpest contribution reframes the central challenge of agentic deployment: governance is solvable; the harder problems are integration complexity and bridging the gap between engineering capability and business process knowledge, which Mantel addresses by embedding business and engineering teams together from the outset.
Our strategic focus areas for the next 12 months
We are prioritising four critical pillars to help organisations advance their AI capabilities:
- Agent Evolution & Learning: We are moving toward “self-healing” agents that continuously evolve by capturing feedback from subject matter experts, customers, and their own failures — including as process knowledge and hyperpersonalisation signals — similar to mature MLOps loops.
- Optimising Unit Economics: We want to drive down the financial cost of building complex agents without compromising engineering quality by using repeatable patterns and agentic coding tools.
- Observability, Governance and ROI Measurement: We are actively exploring how to automatically measure the return on investment (ROI) of production agents to justify and guide our future AI builds, while ensuring responsible AI practices are embedded in how we monitor and manage agents in production.
- Broad vs. Strategic Adoption: We implement “two-speed thinking”, so enterprises don’t delay rolling out general business-facing tools while we build complex, highly specific use cases centrally. As an example, we currently have a team embedded with one large customer focused on end-to-end adoption. This includes an engineering team building custom agents, deployed engineers training and uplifting business teams, and change managers supporting adoption of low-code and no-code options — running in parallel, not in sequence.
Our diverse real-world use cases
We are already past the conceptual phase, deploying both low-code/no-code variants (via Gemini Enterprise) and complex, high-code solutions (built on ADK). Our key applications include:
-
Internal HR agents serving hundreds of thousands of employees.
-
Financial reconciliations and invoice-to-pay automation.
-
Safety and frontline operational support.
-
Customer-facing agentic commerce.
Overcoming Roadblocks & Integration Challenges
-
The Real Bottleneck: While governance is usually solvable and alignment is achievable, our experience shows the harder challenge is managing system integration and dependencies across multiple software surfaces.
-
The Tech-to-Business Gap: Engineering teams who know agents don’t always have the context to reimagine a business process from the ground up, while product owners don’t always have visibility into where the technology is heading.
-
Our Solution: We bridge this gap by embedding business teams directly with our engineering teams. This co-location allows us to solve problems with greater speed, gauge technical feasibility against true business value, and help leadership feel comfortable signing off on risk because they are directly involved in designing the human-in-the-loop mechanisms.
Our Implementation Philosophy
-
Cutting-Edge UI/UX: Upfront, we focus heavily on process mapping, redesign, and user experience design: we’ve created genuinely cutting-edge frameworks for designing effective agentic user interfaces.
-
Capability Uplift: Our engineering experience across some of Australia’s largest and most heavily regulated enterprises means that we can execute at scale in a repeatable way. Throughout the process, we take client teams on the journey with us. Capability uplift happens alongside delivery because we don’t want to leave clients with something they can’t manage or create a dependency on us to keep it running.
“The end state we’re working towards is agents that self-heal and evolve in line with business needs, the way mature MLOps loops do for classical ML models today. Leading organisations are deploying agents across the business at both ends of the spectrum – low-code and no-code variants enabled by Gemini Enterprise, and complex, high-code solutions built on ADK. The conversation has moved from whether to deploy agents, to where to start and how to scale.”
Vihan PatelHead of AI Solutions | Mantel
