There is a new term in the AI lexicon. Paul Hamblin asked Manhattan Associates’ Raphaël Hervé to explain Agentic AI and its potential to transform warehouse process execution.
Artificial Intelligence, Machine Learning, Generative AI – the buzz phrases keep coming. “That’s the world of today, concepts are developed so quickly,” smiles Hervé (pictured, below), Senior Director, Technical and Support Services at supply chain technology leaders Manhattan Associates.
The latest term is Agentic AI. Let’s get straight to the point – what is it?
“If we look at AI in its original definition, for several decades it was about the ability to understand complex algorithms,” he begins. “We then developed IT systems able to make predictions on a very high number of data sets and then even improve those data sets, which we can call ‘Machine Learning’. Then two years ago, Chat GPT arrived along with the phrase ‘Generative AI’, which I would describe as the capacity to make sense of content – whether text, music, sounds, or pictures – and also create this type of content. When you can make sense of language you can begin to ‘push’ these systems to execute tasks for you.
“AI Agents take this a stage further. They are geared towards actually achieving a specific goal, rather than simply making a response.”
Autonomous capability
A key breakthrough is autonomy, he says.
“Operationally, AI Agents are empowered to make decisions and act on those decisions. They also have the ability to interface with the user in normal language. Agents take the instructions in natural language and show the decisions made and steps taken in natural language. Remember, it has an ultimate goal and is able to derive the steps it should follow to reach that goal autonomously.”
As a layman, like many others I’m as nervous of AI’s much-feared potential for chaos as I’m dazzled by its transformationally positive capabilities. Does the autonomy of Agentic AI not make it more likely that repetitive mistakes might become wired into the system?
“Good question, but just like every system it needs to be tuned and optimised,” says Raphaël Hervé. “Let me turn it around. When a complex IT system NOT based on generative AI, or not trained to operate autonomously, makes an error, it is actually very hard to understand why. Because you have to debug, analyse, go into source code. With an agent, you just need to tell it, ‘I think that’s wrong. Why did you say that?’ And the agent will say, ‘I did this for this reason’ and it is therefore far easier to derive the source of the anomaly. Agentic AI is a lot more dynamic in terms of fine tuning than was possible in the past. And unlike your dog or your child it will not resist your instructions,” he adds.
The clarity of AI Agents in explaining what steps they take and why they are taking them is reassuring. “They are very efficient in making adjustments, should you need to,” he promises.
There are several logistics contexts in the Manhattan Associates portfolio of solutions.
An examples is Labor Agent, which is not actually a fully 100% autonomous agent that reaches a goal on its own. Think of it more as an assistant to manage your labour efficiency.
“But it can autonomously sift data, analyse, and procure advice on your labour optimisation,” Raphaël explains. “A typical use case might be a warehouse manager asking Labor Agent if that day’s packing deadlines are likely to be met in terms of human resource. If Labor Agent replies that the team is likely to be late because three people are lacking, it can explore the opportunity to take capacity from elsewhere, for instance from Picking. That team might be able to supply up to five people, so Labor Agent might perhaps select the top three resources with highest ratings and performance on packing. It can then message all parties and reassign via text. The agent is working autonomously and speaking to the user in natural, normal language.”
Time-saving benefits
The question all warehouse managers – and finance leaders – will want answered is, how will we see the benefits manifested in day-to-day use?
One precious win is time, invaluable in any warehouse context.
“The example we just gave is perhaps a 30-second conversation via text, which would have been 15-20 minutes in the past. If packing is late because it lacks three people, it is a complex calculation without the assistance of Labor Agent. You’re looking at process, schedules, performance of packers. There can be big variables, which you then need to compare with what you’re expecting to achieve. The agent can do this for you in seconds.”
Manhattan’s transportation portfolio offers AI Agents with similar benefits.
“Our Freight Invoice Agent is capable of picking up any form of document – PDF, email, spreadsheet – used as carrier invoice materials, and will automatically reconcile actuals with the expected cost of that shipment. This is a role traditionally carried out by manual resources, who spend time receiving documents, comparing screens, shipping costs, taxes, driver hours, and it’s a process that can eat up 15 minutes per invoice. We have built an agent that will automatically ingest anything that comes up, recognise the shipment, align it with expectations, and explain any anomalies. What used to take multiple people hours a day is managed in a few moments.”
Manhattan formally released AI Agents in January of this year, and are marketing the technology to all customer segments, large and small. It even includes Agent Foundry, a developer workspace for customers to build their own agents to their own specifics, either from templates or from scratch.
Raphaël Hervé is brimming with confidence about the prospects.
“We believe Agenti AI is very powerful in terms of productivity gains for our customers. It will drastically improve human-machine interactions, and it will make access to data and functions faster and easier. Customers will enjoy acceleration in project implementation, because integration, mapping, and development are all so much faster.”
