CodeFusion Studio™ 2.0: Accelerating Deployment of Physical Intelligence

For decades, industries have been waiting for AI that can reason in and about the real world. At ADI, we’re making that a reality by delivering advancements in physical intelligence - the ability for AI systems to understand and interact with the electro-physical domains where ADI has spent decades mastering signal conditioning, power, sensing, and actuation.

However, the development and deployment of AI models to embedded systems face many challenges. To realize PI, we need tools that match the complexity of the systems we’re building. That’s where CodeFusion Studio™ 2.0 comes in.

From Embedded Foundations to AI-Enabled Reasoning

Embedding AI models requires more than an Integrated Development Environment in the way technologists traditionally have understood them, it requires platforms that bridge embedded development and AI workflows. It requires the ability to develop on heterogeneous architectures that define the future of edge computing. And more importantly, it must support the kind of agentic, physics-informed intelligence that PI demands.

Let me explain why this matters.

In the past, embedded tools were built for single-core MCUs and deterministic workflows. But PI requires systems that can reason dynamically about thermal properties, magnetic fields, acoustic environments, and more. These aren’t static systems. They’re alive with complexity. And to build them, we need a toolchain that’s just as alive

This is the context that informed the development of CodeFusion Studio 2.0. It supports multicore debug, system-level planning, and AI model integration, all in a unified workspace. It’s Zephyr-first and open. That means reproducibility, automation, and extensibility are baked in. For PI, this is critical. We’re building agents that reason at the edge, and those agents need to be trained, deployed, and debugged in environments that reflect the real world.

AI Workflows That Meet the Edge Where It Is

One of the most critical features of CodeFusion Studio 2.0 is its end-to-end AI pipeline. Developers can import models from TensorFlow or PyTorch and generate inference-ready code in minutes. With the Zephyr AI Profiler, they can monitor latency and memory all without touching hardware.

This is a game-changer for PI. Our goal is to embed intelligence directly into products, whether it’s context-aware audio in hearables or adaptive control in robotics. CodeFusion Studio makes that possible. It turns AI from a bolt-on feature into a core design principle.

In addition to inference, the platform supports AutoML for Embedded, enabling dataset training and optimization within the same workflow. That means our agents can learn from the physical world, adapt to it, and act within it, all while staying within the constraints of edge hardware.