The agentic AI awakening: inside the tech powering autonomous innovation

Agentic AI is getting ready for primetime, and not just because the tech world loves a new buzzword. The latest agents are goal-driven, context-aware and increasingly capable of taking on entire tasks that used to require complex workflow engineering.

As more people get comfortable chatting with AI, they’re starting to wonder: if it can write an email, why can’t it just handle the whole task for me?

Agentic AI is the step from intent to outcome. You simply describe the goal in natural language, and the agent orchestrates the steps required.

In industries like automotive and logistics, that potential is especially powerful. To take a closer look at the layers behind how these agents work, we sat down with Aleksandra Kovacevic, Senior Director of Responsible AI at HERE.

From perception to collaboration, agentic AI relies on layered intelligence to power smarter, safer and more autonomous systems.

The perception layer

Before an agent can decide or act, it has to understand what’s going on around it. Think of the perception layer as its senses: continuously capturing, validating and aligning signals about what’s happening right now. It pulls in real-time, trustworthy information and turns it into a clear picture of what’s happening right now. And it goes far beyond simply spotting where things are. It needs to grasp the whole situation.

In logistics, for example, this means monitoring traffic patterns, road closures, vehicle health, weather conditions, access rules and anything else that could shape the journey ahead. It’s not just about knowing where things are; it’s about grasping the full context of the situation.

“Everything starts with a reliable, high-quality map,” explained Kovacevic. “If the agent doesn’t understand the world around it accurately and in real time, everything that follows is guesswork.”

HERE’s map serves as this shared source of truth that is continuously refreshed, enriched with private datasets.

The learning layer

The learning layer is where the agent becomes more useful over time. Every decision and its outcome are analyzed, refining the agent’s anticipation, preferences and reasoning capabilities. If a recommended café is consistently ignored by users or a suggested route proves inefficient, the agent learns and adjusts its future recommendations.

“The challenge is to be able to trust AI to act on our behalf,” said Kovacevic. “Trust comes from understanding the “why” behind a decision. If a truck is rerouted, the agent needs to know the reason so it can negotiate the most optimal decision”.

Every driver correction, every rejected recommendation, every surprising outcome becomes training signal. In practice, this means the map itself improves; wrong entrances get corrected, better routes surface and new patterns emerge.

This constant refinement is how agents become more predictable, safe and trustworthy with every task it performs.

Read more — Mixing maps with mindfulness: what agentic AI means for automotive and logistics

The decision layer

The decision layer is essentially the system’s “gut instinct,” the place where raw information gets mixed with experience to produce smarter choices. Think of the difference between a new driver and a seasoned one: a beginner only responds to what’s directly in front of them, while an experienced driver slows for a blind bend because they remember how those situations usually play out.

That’s where HERE has a real advantage. “HERE has 40 years of experience. Four decades of historical data,” said Kovacevic. “You can imagine how much memory and anticipation we can build on top of that.”

With this depth of data, the map stops being just lines and shapes. Every stretch of road becomes a story: what traffic typically looks like, where accidents tend to happen and how different vehicles handle the terrain and how conditions shift by time of day.

By combining memory and insight, the agent makes decisions that aren’t just faster but also safer, wiser and more aligned with how humans operate.

The collaboration layer

The future of automation isn’t about a single agent working on its own, but a network of agents collaborating to optimize entire systems. A vehicle agent will need to communicate with a warehouse agent, which in turn speaks to a supply chain agent.

For this to work, agents must not only exchange results but also the reasoning behind those results.

“Explainability is what enables negotiation between humans and agents and between agents themselves. That is the foundation of trust,” said Kovacevic.

When an agent recommends Route B over a shorter Route A, it must be able to explain that it did so to avoid a bridge your truck can’t pass. This transparency allows an operations manager to understand the trade-offs, stay in control and negotiate better outcomes. As these systems evolve, agents will use this same logic to negotiate with each other, finding the optimal delivery slot or rerouting an entire fleet in response to a disruption.

Building the agentic future, layer by layer

Agentic AI could help to make our supply chains more resilient, our vehicles smarter and our operations more efficient. But it can’t run on general-purpose models alone.

By providing the foundational data for perception, the historical context for decision-making, the feedback loops for learning, and the framework for collaboration, HERE is supplying the critical location intelligence needed at every layer.

“In the agentic era, the map becomes something agents can reason with not just look up. And once you give agents grounding, memory and explainability, they can act reliably,” said Kovacevic.

For automakers and logistics companies, this multi-layered approach is the key to unlocking the full potential of autonomous systems and building a smarter, more connected world.