What is agentic AI, and how is it used in logistics?

Agentic AI is moving quickly into logistics. This article explains what it is, why data matters so much, and what organizations should establish first. It draws on the whitepaper Agentic AI in Logistics: From Data Infrastructure to Operational Intelligence, co-authored by Smart Freight Centre, SINE Foundation, Way Data Technologies and Pathfinder.

What is agentic AI?

Agentic AI is built on three concepts:
  • A language model. This is a statistics-based model trained on vast amounts of existing data. Large language models (LLMs) excel at handling sequences of words and capturing patterns in text. The model lets the agent process input and generate output.
  • System instructions. This is fixed text the model receives before any interaction. It defines the agent’s role, its domain knowledge and how it should behave. For example, a fleet management agent can be told to know EU driver hours rules and to flag compliance risks.
  • Tools. A tool is a specific action the agent can take, backed by software it doesn’t run itself, such as an API, a routing engine or a database query. Give an agent “schedule_maintenance” and it can book a maintenance slot.
Tools are what separate an agent from a chatbot. A chatbot produces text for a human to act on. An agent acts in the systems around it.

Why now?

Gartner forecasts that supply chain management (SCM) software with agentic AI capabilities will grow from less than $2 billion in 2025 to $53 billion in spend by 2030. It also predicts that by 2030, 60% of enterprises using SCM software will have adopted agentic AI features, up from 5% in 2025. DHL’s Logistics Trend Radar 8.0 identifies agentic AI as one of its most impactful trends.

Companies are already acting on this. Nvidia and Kawasaki Heavy Industries announced a joint effort to build a next-generation digital shipyard in Japan, co-developing AI-powered robots for shipbuilding. Descartes  showcased AI innovations at its 2026 Innovation Forum, including AI agents that take on freight visibility follow-up.

Yet, the whitepaper finds that the gap between expectation and operational reality remains wide. Gartner similarly expects that as Supply Chain Management vendors make AI capabilities widely available, many companies will be slower in deploying them due to their processes, data, governance, and operating models not being ready to fully support the technology.

Why data defines success

The whitepaper’s central argument is that whether an AI deployment succeeds or fails depends on the underlying data: its availability, its quality, and the processes and standards governing it.

The key concept here is data asymmetry: one party has more or better data than another. In logistics, this is a structural problem:
  • Shippers and logistics service providers see shipment-level data, such as origins, destinations, volumes and timing.
  • Carriers and fleet operators hold execution data from connected vehicles, such as energy consumption, charging durations and vehicle health.
  • Charging infrastructure data is scattered across dozens of operator apps and platforms, often unverified and inconsistent.
When real operational data is missing, planning tools fall back on industry averages and add safety margins. Over five to ten years, those margins can compound into significant over- or under-investment.

Agentic AI raises the stakes. An agent working on incomplete data may fail visibly or, more insidiously, hallucinate. Where humans might recognize uncertainty and pause, agents can turn isolated data gaps into systematic decision errors at speed and scale.

Five common misconceptions

  • “AI will replace software.” For decisions like fleet procurement or charging investment, purpose-built simulation and optimization tools remain essential. The right model is AI with specialized tools.
  • “AI can fix poor data.” AI can help structure messy data, but it cannot manufacture accuracy. An agent that confidently generates an inaccurate ETA could make an unfounded routing recommendation. “Garbage in, garbage out” still applies.
  • “AI generates new knowledge.” Language models transform existing information; they don’t create new facts. When data is missing, a model may fill the gap with a confident-sounding fabrication.
  • “AI is only for software engineers.” A fleet manager, a sustainability officer or a procurement lead can all be supported by an agent configured with the right data and tools.
  • “Agentic AI makes standards obsolete.” Shared definitions matter more as agents exchange data. AI lowers the cost of adopting standards, not their value.

Where does agentic AI create value in logistics?

  • Tender analysis. AI can clean large datasets, visualize flows and identify round-trip opportunities far faster than manual analysis.
  • Fleet maintenance. An agent can review telematics data overnight, flag anomalies and draft a prioritized action list for the fleet manager to approve in the morning.
  • Electrification planning. When the right data is available, agents can screen lanes for electrification, compare scenarios and evaluate charging implications.
According to the authors, the most reliable model today is AI-accelerated human planning: agents do the analytical heavy lifting, and people make the final call on consequential decisions. For organizations that are keen to start deploying Agentic AI, the whitepaper offers some useful steps to consider: 

1. Before deployment, such as identifying data gaps and assessing governance. 
2. During deployment, like establishing human "check-points" and adopting open standards such as iLEAP. iLEAP, is an open standard for exchanging logistics emissions data, developed by SFC, SINE Foundation and contributions from many members and industry partners.
3. When scaling Agentic AI systems, such as up-skilling personnel and contributing to shared data ecosystems.

Key takeaways
  • AI is as good as the data it has access to, the governance around it, and the standards that enable data access.
  • AI can’t fix poor data or replace purpose-built logistics software.
  • Today, agentic AI works best as an accelerator for human decisions, not a replacement for them.
Read the full whitepaper: Agentic AI in Logistics: From Data Infrastructure to Operational Intelligence

Courses related to this subject: