As always, things move so fast in the AI space (no one can hear you scream) that you no longer know if you’re coming, going, or doing something else entirely. For example, when generative AI came out, I was barely able to wrap my head around perceptual AI, and when agent AI showed up with a metaphorical flugelhorn fanfare (much louder than you’d expect from a metaphorical fanfare), I was just beginning to understand generative AI.
I think it’s safe to say that most people believe that more advanced forms of AI (generative and agentic) only exist in the cloud. But when it comes to perceptual AI, generative and agentic AI applications are increasingly moving to the edge, where the “rubber of the internet” intersects the “roads of the real world.”
One company dedicated to bringing artificial intelligence from the cloud to the real world is NXP. Its name is an abbreviation of “.”NexternalXPThe word “Elliens” is becoming more and more appropriate as the days go by. Rather than chasing eye-popping TOPS numbers or boasting data center bravado, NXP has focused on something far more practical: providing an edge AI platform that engineers can deploy in the real world. NXP’s processors span microcontroller units (MCUs), application processors (APs), and disparate SoCs to right-size AI acceleration and real-time control, vision, and this is AI designed for factories, vehicles, machines, and infrastructure, not press releases.
NXP is focused on AI at the edge (Source: NXP)
Equally important, NXP is treating edge AI as a full-stack problem. Silicon is combined with mature software, tools, and frameworks that allow developers to move from model to production without heroic efforts. This, combined with NXP’s extensive experience in security, safety and system reliability, results in an edge AI strategy that is firmly rooted in reality. Against this backdrop, the company’s recent CES announcement about agent AI doesn’t feel like a sudden change of direction or a grab at the buzzword. Instead, they feel like:
The next logical step in the journey NXP has been methodically preparing for some time.
But I fear we are getting ahead of ourselves. Let’s quickly set the scene by remembering that cognitive AI refers to recognizing what’s going on in the world, such as computer vision (object detection and identification) and speech recognition. This can be summarized as “sense” as shown below.
Feel, think, act (Source: NXP)
NXP was one of the first companies to launch the i.MX 8M Plus in 2020, a general-purpose embedded processor with a dedicated AI accelerator. This allowed developers to switch from traditional computer vision techniques to AI for tasks such as object detection and classification. For example, developers no longer need to explicitly define feature extraction. Instead, you can just say, “This is a cat and this is a dog,” and the AI will learn what cats and dogs are and perform feature extraction on its own. At that time, the cycle for a model that stopped working was about 3 years.
Get edge-friendly and high-performance on edge devices, anywhere but in the cloud.
With the advent of generative AI and large-scale language models (LLMs) like ChatGPT, tasks such as content, image, and code generation are now possible. This can be thought of as “sensing and thinking.” Again, this started with the cloud. NXP officials quickly focused on making generative AI relevant and practical at the edge. It was about a two-year cycle to have high-performance, edge-profile versions of these models.
We now have agent-based AI that can be thought of as “sense, think, and act.” In this case, it took NXP personnel less than a year to move from what was announced and launched in the cloud to a reality at the edge.
The reason I was thinking about all this was because I was just chatting with Ali Osman Ors, Global Director of AI Technology and Strategy at NXP. As Ali said about agent AI, “Last year, everyone was talking about this AI. This year, we demonstrated it on the CES show floor and showed how to create and deploy agent systems that can perform autonomous actions.”
Before we proceed, let’s take a quick look at the hardware and software components of NXP’s AI platform prior to CES 2026, as shown below (note that eIQ stands for “Edge Intelligence”).
Intelligent edge systems require the latest and greatest HW, SW, and AI ecosystem (Source: NXP)
Let’s start with NXP’s hardware portfolio. NXP’s hardware portfolio is expanding every day. In addition to the advanced MCX MCU, the even more advanced i.MX RT crossover MCU is also available. Not only are they extremely power efficient, but they also have a wealth of features and performance (think smart glasses, smart watches, etc.). Then there are i.MX APs, which use higher performance multi-core CPUs, GPUs, DSPs, and NPUs to provide heterogeneous computing.
In each of these categories, NXP offers devices and families with integrated dedicated AI accelerators (NPUs). They initially used third-party NPU IP. We recently transitioned to using eIQ Neutron NPU cores that we developed in-house.
But wait. Because NXP acquired Kinara in October 2025 along with the amazing Ara AI processor. NXP currently offers what is currently called the Neutron GT NPU in a standalone Ara DNPU device, but it won’t be long until this NPU IP core is available on a new family of MCX MCUs, i.MX RT MCUs, and i.MX APs (I don’t know about you, but my brain is wobbling on gimbals).
On the software side, the eIQ AI Toolkit is NXP’s core software foundation for deploying AI models to edge devices. Its primary role is to optimize models originally developed for cloud or desktop environments to built-in constraints (e.g., limited memory, compute, power, etc.).
eIQ Time Series Studio is an automated machine learning tool with a special focus on time series data such as sensor streams, telemetry, and industrial signals. eIQ Model Creator targets a wide class of ML problems beyond time series, providing an automatic path from raw data to optimized edge-aware models. eIQ GenAI Flow also extends NXP’s software stack into generative AI with a focus on efficiently running transformer-based models on edge hardware.
But these aren’t what I really wanted to talk about.
At CES 2026, NXP stopped short of introducing a standalone “agent flow” and instead announced a more ambitious product, the eIQ Agentic AI framework (I think it’s easier to imagine drum roll at this point). In contrast to being a peer to the eIQ GenAI flow, the eIQ Agentic AI framework can be visualized as sitting “on top” and coordinating multiple flows. include eIQ GenAI flow.
Simply put, the eIQ Agentic AI framework provides the orchestration layer needed to build autonomous edge systems that coordinate multiple AI models, tools, and actions in real time. It integrates naturally with existing elements of the eIQ stack, such as eIQ GenAI Flow, AI toolkits, and automated ML tools, allowing perception, inference, and control to work together on the device. The result is not only generative AI at the edge, but also agent-driven systems that can sense, decide, and act without relying on continuous cloud connectivity.
Ali described one of the CES demos of the eiQ Agentic AI Framework in action, as shown below (MCP stands for Model Context Protocol, an open standard that allows LLMs to communicate with external data sources, applications, and services).
Demonstration of NXP’s eiQ Agentic AI framework in action (Source: NXP)
In this case, the agent edge AI system was assembled from disparate devices working together as a single autonomous whole. The i.MX 8M Plus AP acts as the front end of the system and handles multimodal inputs such as video, audio, and text. A separate Ara-2 NPU ran large language and vision models, and a small MCX microcontroller handled simple real-time motor control, not as an “intelligent” device in its own right, but as a controllable tool exposed to an AI agent.
The system was given a small set of tools to analyze video scenes, make outgoing calls and messages (including via WhatsApp), and start motors to emulate responses such as sprinklers and HVAC vents. He was then tasked with monitoring the video feed of an industrial environment, with instructions only to “mitigate the impact and notify supervisors if an incident occurs.”
Importantly, there are no explicit rules or if-then logic defined. Instead, visual and language models running on Ara NPUs interpreted the scene, determined whether something constituted an “incident,” autonomously decided which tools to invoke, and triggered physical actions and notifications as needed. The result is a fully on-device, agent-driven system that senses, reasons, and acts without being explicitly programmed for a specific scenario.
I have to admit I’m amazed at how things are going and how far we’ve come in such a short amount of time. One thing that really “ticked my fancy,” so to speak, was when Ali said that the video stream of the “incident” fed into the agent AI system was itself generated by the AI.
I can’t help but think how strange and frankly unbelievable all of this would have seemed just a few years ago. Autonomous systems that watch videos, interpret events, generate code, make decisions, and perform real actions, all on a device without being explicitly programmed for a specific scenario, would once reside firmly in the realm of science fiction. They are currently being demonstrated on the trade show floor using commercially available hardware and software.
Whether all of this excites you, makes you nervous, or both, one thing is for sure: Edge no longer just senses the world. We understand that and we are starting to act on it. We certainly live in interesting times (I hope they don’t get too interesting).
related
