Smart mines get smarter as agentic AI steps up to connect data boost productivity and put intelligent decision making in every miner’s toolkit

The future of mining is already here - and it’s being shaped by AI systems that can think, act and integrate seamlessly with the tools you already use.

The Promise and Practice of Digital Intelligence for Mine Productivity – Foundations for Agentic AI Success was the title of the keynote delivered by Ben Cabanas, APJ Technology Director at Amazon Web Services (AWS), during the APCOM2025 conference in Perth. In a fast-paced and example-rich presentation, Ben unpacked the unprecedented pace of AI development, explained how “agentic” AI can revolutionise operations, and outlined why mining companies are in a prime position to adopt it now.

AI’s breakout moment

“AI is not new,” Ben began. “But what we’re seeing in 2025 is fundamentally different to anything before. We’ve moved beyond chatbots to AI that can reason, explain its thinking and complete tasks that previously required constant human oversight.”

That leap, he said, can be traced back to the 2017 invention of the transformer architecture, outlined in the landmark paper Attention is All You Need. This breakthrough underpins the large language models that dominate AI development today. Since then, the field has evolved beyond text, with multimodal AI systems now capable of processing and generating images, video, and multiple content types simultaneously.

These advances aren’t just about capability - they’re about accessibility. “The cost to generate a million tokens at GPT-3.5 quality has dropped from around $20 in 2022 to just a few cents today,” Ben noted. “That democratises AI, putting it within reach of smaller mining companies, startups and developers who previously couldn’t afford to experiment.”

Speed and cost – the great equalisers

Ben explained that AI is now not only more powerful but also much faster to run. Inference - the process of using a trained AI model to generate an answer or perform a task - has improved elevenfold in just a few years thanks to better architectures and more efficient infrastructure.

“This matters because speed and cost are the great equalisers,” he said. “When AI can complete complex tasks in seconds and for a fraction of a cent, the barriers to adoption in industries like mining virtually disappear.”

Bringing AI into the mining value chain

For mining, the real transformation lies in embedding AI directly into existing workflows. Ben cited examples already being deployed:

• Exploration – AI-powered computer vision automates drill core analysis, detecting trace minerals and mapping ore zones faster and with greater accuracy.
• Mine planning – Machine learning models inform financial risk assessments, improving decision confidence.
• Production – IoT sensors feed AI systems that optimise throughput and energy usage in real time.
• Maintenance – Predictive models flag potential equipment failures before they occur, allowing for targeted interventions and reduced downtime.
• Remote operations – Digital twins of facilities allow engineers to “walk” through a plant virtually, viewing real-time data and even triggering operations remotely.

“It’s about integrating AI with your operational data so it can work with the systems you already have,” Ben said.

From assistants to agents – the agentic AI revolution

One of the most compelling ideas in Ben’s keynote was the concept of agentic AI - fully autonomous, multi-agent systems that can reason, collaborate, and complete complex workflows with minimal human supervision.

“We’ve gone from assistants that follow simple rules, to agents that automate whole workflows, to agentic systems that operate with human-like logic,” he explained. “These agents can break down tasks, adapt to context, and learn over time.”

A breakthrough in enabling this shift is the Model Context Protocol (MCP), which Ben likened to “a USB for AI.” MCP is a standardised interface that allows AI agents to connect seamlessly with a range of platforms without the need for custom integration. “Before MCP, every integration required unique APIs and development work,” he said. “Now a single standard lets AI interact with anything from SAP to geological modelling software.”

Connectivity without limits – Project Kuiper

For mining, especially in remote regions, connectivity has always been a barrier to implementing advanced digital systems. Ben’s answer to this: Project Kuiper, Amazon’s low-Earth orbit (LEO) satellite network.

“We’re launching 3,236 satellites to deliver high-speed, low-latency internet anywhere on Earth,” he said. “By mid-2026, half the constellation will be active, enabling secure, real-time AI-powered digital twins even at the most isolated mine sites.”

With latency of just 30–50 milliseconds, Project Kuiper will provide fibre-like performance for remote operations - and, crucially, will connect directly into AWS cloud environments for secure data management and AI deployment.

Lessons from Amazon’s own playbook

Ben illustrated the transformative potential of AI with examples from Amazon’s global operations. Since 2012, the company has deployed over 750,000 AI-powered robots in its fulfilment centres. These robots don’t just move items - they coordinate with humans, optimise picking routes, and adapt to changing workflows.

When real-world training data is scarce, synthetic environments take over. Using NVIDIA’s Isaac Sim platform, Amazon trains its robots in photorealistic simulations, improving accuracy and reducing development time from months to days.

The same simulation techniques, Ben argued, could be applied to mine-site robotics, autonomous haulage, and plant equipment optimisation.

AI that modernises legacy systems

Beyond new deployments, Ben stressed AI’s value in working with and modernising legacy systems. At Amazon, AI agents have helped overhaul more than 10,000 internal applications, saving an estimated 4,500 developer years and over $260 million in operational costs.

“That’s the scale of efficiency gains mining companies can aim for by connecting their data, systems and AI capabilities,” he said. “It’s not about ripping out everything you have - it’s about making what you have smarter.”

Why the time is now

In closing, Ben delivered a clear message to mining leaders: the industry already has much of the infrastructure needed to capitalise on AI. Years of investment in automation, digital twins, and operational technology have laid the groundwork.

“The mining industry has invested heavily in automation, optimisation and digital twins,” he said. “By liberating and connecting those data assets, ensuring secure connectivity, and applying agentic AI, you can unlock benefits that were previously not possible.”

His final takeaway was both a challenge and an invitation: “The future of mining is smart, connected and AI-powered - and it’s available today.”