Embedded Hardware Development with Agentic AI

Some people think they are doing AI because they have Copilot enabled in their editor. But there is a long way from advanced autocomplete to a workflow where you kick off a task, walk away from the machine, and come back to working code, passing tests, and a commit.

That difference is what this evening is about — and we take it all the way down into the hardware, where things cannot merely look right. Either the LED blinks, or it does not.

Christian Dalager, principal consultant at Eksponent, posted on LinkedIn this summer about "Eksponent MØDER": a meeting recorder that records, stores, syncs to a backend, transcribes with an open-weight Whisper model, summarises with a local LLM, and syncs the result back to the device. The hardware cost DKK 258. It took around ten hours. The design was done in Claude Design, and the low-level firmware in C and LVGL was written with Claude Code on a standard ESP32-S3 controller.

Christian will talk about how the project actually went — where it flew, where it stalled, and what he would do differently. And, just as importantly, what it means for the way we plan and price development work when the cost of technical implementation drops this drastically. The point is implication as much as implementation.

Jørgen Kragh Jakobsen has spent the past year working with agentic workflows in chip and embedded hardware development — from RTL and verification through to bring-up on real boards. He gives a concrete "how to get started": how to break a large task into something an agent can actually work on, how to make it verify its own work against hardware rather than against its own assumptions, how to set up context and a rule set for a repository, and what to do when it goes off the rails.

And then we play. Every participant is handed an ESP32-S3 module with a round 360×360 touch display, a rotary knob, a microphone, a speaker, and an LED ring — the same kind of hardware Christian's device is built on. You get a set task and a starter repository, and you spend the rest of the evening building your own firmware with whatever agentic setup you prefer. We close with a demo round where everyone shows how far they got.

The module is yours, and you take it home.

What you take away

A working ESP32-S3 module with a display, audio in and audio out — with your own firmware on it

A concrete picture of what an agentic development flow looks like in practice

A way of working that applies just as well to pure software projects as to embedded

A starter repository with project setup, agent rule set and toolchain to build on

Who it is for

Embedded developers, software developers, architects, tech leads and data scientists who are curious about what agentic workflows can do when they are used in earnest. No prior experience with ESP32 or LVGL is required. You should be comfortable working in a terminal and with flashing firmware onto a board.

The workshop is hosted jointly by IDA Embedded and IDA AI, so we get both the embedded crowd and the data science crowd in the same room. That is very much the point.

Sandwiches will be served during the workshop.

Language: the evening is held in English.