Why Mantel thinks 2026 is shaping up as the tipping point for agentic AI
Partners must help guide organisations from pilot to production as agentic rewrites the way we work.
Organisations are poised to make the leap from agentic AI pilots into stable, real-world business use cases.
“By 2026, the primary focus will be on productionising agentic AI,” Adam Durbin, chief technology officer, Mantel explained.
But they won’t do it on their own. They’ll need to be guided by experienced partners harnessing new tools and solutions.
“It’s quite easy to prototype and get something working but the focus must be getting the reliability and consistency out of agentic systems,” Durbin said speaking at this week’s AWS re:Invent conference in Las Vegas.
Mantel, which has just signed a three-year strategic collaboration agreement (SCA) with AWS, is looking to deepen its AI capabilities and intensify efforts to develop customer solutions.
The business has more than 500 AWS certifications across its team and will be focusing on developing utilising agentic tools such as Amazon Bedrock, Amazon Q and Amazon SageMaker.
“We’re really looking forward to autonomous systems and agentic — that’s the focus for us moving forward,” he told CRN Australia.
However, Durbin highlights a number of complex technical challenges they must overcome for organisations to achieve reliability and scale of agentic workflows.
Evaluations are the critical first step to assess whether the agentic AI behaves as expected across various scenarios, not just during initial development.
“It’s easy to get systems that work 80 percent of the time,” he said.
Most organisations also need to modernise legacy tools and workflows to fit with new agentic AI architectures. Then there’s the question of choosing the right AI model is also crucial for balancing cost, accuracy and processing speed.
“Token management is complex, and high costs can come with using more powerful models, so optimising token usage and choosing lighter models for some tasks becomes important.”
With this in mind, Amazon has introduced lighter models with specific use cases to reduce costs and increase speed where it’s needed.
“The economics need to stack up to build an agentic solution and availability of models on the platform is really important.”
Performance and latency management need to be addressed while creating new processes to handle maintenance, updates and monitoring long-term, not just initial launches.
“That’s where the experience of partners comes in because we’re seeing brand new styles of building systems … and taking it all the way through the production,” he said.
With a raft of new agentic tools and solutions announced by AWS, Durbin expects Mantel’s agentic engagements to grow.
“Every single one of our projects now looks at using a genetic AI coding style, whether that's a greenfield build or tech modernisation. It's just embedded in the ways of working.”
Rosalyn Page travelled to AWS Re:Invent 2025 as a guest of AWS.