Agentic AI in food: overview
- Agentic AI helps retailers and manufacturers improve operational efficiency
- AI agents balance demand production, capacity inventory and procurement
- Autonomous actions accelerate decisions but increase governance and oversight risks
- Poorly managed agents may amplify errors across business functions
- Human judgement remains essential to prevent overreliance and fraud risks
It’s only been a few years since the launch of ChatGPT, but chatbots are already ubiquitous.
As AI systems develop, another form of AI is gaining prominence: agentic AI.
Agentic AI systems, which can act autonomously, could help both manufacturers and retailers speed up decision-making and improve the efficiency of operations. But there are, of course, risks.
How agentic AI can boost decision-making
Agentic AI is already used extensively by consumers in commerce. But it can also make operations more efficient from the perspective of retailers and manufacturers themselves.
In retail, AI agents can help managers anticipate demand on a store-by-store basis, says Nick Diamond, global lead and CEO advisory for consumer goods at consultancy Accenture. It can help them adjust product replenishments and markdowns (price reductions) as products get closer to their expiration date.
It can also enable retailers to respond more quickly to changes in weather and local buying patterns, helping to reduce food waste and improve availability.
“The important distinction from earlier automation is that agents don’t just analyse or recommend: they can orchestrate work and act, with humans setting the objectives and guardrails.”
Nick Diamond, global lead and CEO advisory for consumer goods at Accenture
The use of agentic AI for manufacturers is similar. “Agents can help continuously balance demand, production capacity, ingredients and inventory,” he says.
For example, they can identify possible supply disruptions and their potential impact on production schedules, assess alternative ingredients or suppliers, and even recommend the best response.
In short, they can speed up the decision-making process. “Food businesses make thousands of decisions every day: what to make, where to put it, how much inventory to hold, what price to charge and what to promote.
“Agents can continuously interpret data and coordinate decisions across those activities rather than relying on sequential human processes. The important distinction from earlier automation is that agents don’t just analyse or recommend: they can orchestrate work and act, with humans setting the objectives and guardrails.”
Finally, simply having access to an AI agent does not give food companies a competitive advantage, argues Diamond, as these agents will soon be widely accessible.
The advantage could instead come from how quickly they redesign processes around AI, what proprietary data they can bring to the table, and how quickly they can turn this insight into action.
The risks of agentic AI
Nevertheless, it is precisely what makes agentic AI so useful that poses the greatest risk. It does not advise; it acts.
“An inaccurate recommendation from traditional AI can be reviewed before anything happens; an autonomous agent could potentially propagate that decision across pricing, purchasing, inventory or customer interactions at speed,” says Diamond.
This makes elements such as transparency, governance, data quality, and human oversight even more important, he says. Humans should not be removed from the equation completely: they must decide carefully which decisions should be made autonomously.
There is also a risk that agentic AI could distance managers from the day-to-day operations in their business, admits Diamond, as more operational decisions will be automated.
Nevertheless, this can be managed. A good agentic AI should simply reduce the time a manager must spend on assembling the information, and instead increase the time they spend making judgement calls and addressing issues that matter most.
“The objective should be better management visibility, not management abdication,” says Diamond.
Fraud is also a risk. While retailers should make their websites discoverable by AI agents, there is always a risk that these agents are aiming to commit fraud, explains Andrew O’Connor, agentic AI lead at consultancy PSE Consulting.
“There is a little bit of a challenge of understanding, from an agent perspective, what is a good agent, what’s not trying to commit fraud or do anything malicious versus the flipside on that.”
While AI increases the capacity to do things quickly, it also enables fraudsters to find new ways around the systems in place to stop them.
Fraudulent activity “increasingly arrives as automated traffic that looks like a genuine shopping agent.”
In short, AI agents can boost the efficiency of a business, helping it make quicker decisions and heavily streamline cumbersome processes.
Yet the autonomy of the agents means that their users must be careful that human oversight is retained, otherwise a mistake could go much further.
