The State of AI in Agrifood: Insights from Yield by Artesian

State of Play on AI: What the Last 12 Months Have Changed for Agrifood

As we approach the end of 2025, it feels like an appropriate moment to pause and take stock of how far AI has progressed — not in theory, but in practice.

At Artesian’s inaugural Yield agrifood summit, two sessions in particular crystallised this shift. Dawid Naude (Pathfindr) and Vinod Bijlani (Hewlett Packard Enterprise) offered complementary perspectives on where AI is heading, what has materially changed over the past 12 months, and what this means for real-economy sectors like agriculture.

One observation from Vinod captured the moment perfectly: for much of the past decade, AI speakers were often placed at the end of conference agendas. Today, they are first — and increasingly central to every strategic conversation. That change reflects a deeper reality: AI has moved from experimentation to infrastructure, and from novelty to necessity.

1. From Chatbots to “Thinking Systems” and AI Agents

The first major shift discussed at Yield was the evolution from simple chat interfaces to agentic, reasoning systems.

Rather than just responding to prompts, these newer models can plan, sequence tasks, and take action across systems. In an agrifood context, that means AI moving from passive insight to active operational support. Examples discussed included systems that continuously monitor farm operations and recommend interventions in real time, and platforms that fuse text, imagery, sensor data and geospatial information into a single reasoning loop.

The implication is significant. AI is no longer just “smart autocomplete” or decision support layered onto existing workflows. It is increasingly behaving like a co-worker — augmenting agronomists, operators and managers rather than simply informing them.

That shift brings opportunity, but also responsibility. If AI systems are acting, not just advising, questions around skills, governance, accountability and trust become just as important as model accuracy.

2. Infrastructure Is Strategy: AI Factories, Compute and Scale

A core message from Vinod Bijlani’s session was that AI is no longer primarily an algorithm problem — it is an infrastructure problem.

Organisations that succeed with AI will not simply be those with clever models, but those with access to the full stack required to deploy AI reliably and repeatedly. That includes high-performance compute such as GPUs and accelerators, secure and governed data pipelines, and what Vinod described as “AI factories”: industrialised, repeatable systems for building, deploying, monitoring and improving AI at scale.

This matters enormously for agrifood. Agricultural AI is not built on static datasets; it depends on streaming data from farms, supply chains, logistics networks, weather systems and markets. Models must be retrained, monitored and adapted continuously as conditions change.

Over the past year, this reality has started to reshape investment and policy decisions in Australia. Significant capital is now flowing into domestic AI data centres and sovereign AI factories designed to give local organisations access to serious compute onshore. The message from Yield was clear: in 2025, infrastructure is AI strategy.

If agrifood organisations do not know where their models run, where their data resides, or who ultimately controls that stack, they are operating with a strategic blind spot — particularly in a sector that underpins food security and export competitiveness.

3. Sovereign AI Is an Agrifood Imperative

At Yield, sovereign AI was not discussed as an abstract national ambition, but as a practical requirement for agriculture.

Agrifood is uniquely exposed. It operates under biological variability, climate risk and long production cycles. It relies on infrastructure-heavy supply chains and is deeply embedded in global commodity markets and trade settings. These characteristics make agriculture both data-rich and risk-sensitive — exactly the type of sector where sovereignty, governance and resilience matter.

For agriculture, sovereign AI capability delivers three concrete benefits to growers and producers. First, relevance: models trained on Australian soils, climates, pests, diseases, water constraints and regulatory settings will always outperform generic global models.

Second, trust and control: growers need clarity on who owns farm data, how it is used, and under what rules — particularly as AI systems move closer to operational decision-making. Third, resilience: critical industries cannot be entirely dependent on offshore compute, governance frameworks or commercial incentives that may not align with local priorities.

Sovereign AI is not about isolation. It is about ensuring Australia can participate in global AI ecosystems on its own terms, while protecting the long-term interests of growers and the industries that support them.

4. Agriculture as the Testbed for Responsible AI

A recurring conclusion across both sessions was that agriculture is the ideal testbed for responsible, sovereign AI.

It is complex, safety-critical and economically material. Success requires cooperation across growers, RDCs, corporates, researchers, startups and investors. It demands shared trials, trusted data, and clear governance frameworks that balance innovation with accountability.

If AI can be deployed effectively in agriculture — with shared risk, transparent data use, local infrastructure and real on-farm outcomes — it creates a blueprint that can be applied to other critical sectors such as energy, health, logistics and defence.

For growers, the upside is tangible: better decisions, reduced risk, improved productivity and greater resilience in the face of volatility. For the broader system, it demonstrates how AI can be embedded responsibly in the physical economy, not just digital services.

So Where to From Here?

The combined takeaway from Dawid Naude and Vinod Bijlani’s sessions was unmistakable: AI is no longer a discrete technology layer. It now sits across every stack, every sector and every strategic decision. In agrifood, this shift is particularly consequential — not because AI replaces human judgement, but because it augments it in environments defined by complexity, risk and biological uncertainty.

The real strategic choice is no longer whether to adopt AI, but whether Australia builds the capability, infrastructure and governance to deploy it on our own terms. That means leaders moving beyond delegation and developing a working understanding of AI’s possibilities and limits. It means organisations identifying where AI can create tangible value today — on-farm, across supply chains, and within RD&E — rather than deferring impact to some future horizon. And it means coordinated action between growers, RDCs, corporates, technology providers, investors and government to ensure data, compute and models are trusted, governed and sovereign.

What was most encouraging at Yield was that many of these foundations are already in place. World-class growers are engaging directly with new tools. RDCs are expanding traditional RD&E into distributed, startup-driven innovation. Corporates are providing scale, validation and later-stage capital. Investors are increasingly focused on long-term, strategic outcomes, not just short-term returns.

The opportunity now is alignment. If these elements move in step — with agriculture as both beneficiary and testbed — Australia has the chance not only to apply AI effectively in agrifood, but to build a blueprint for how responsible, sovereign AI can strengthen critical industries more broadly. Yield made it clear: the pieces are there. What comes next is how deliberately we bring them together.