Embedded development has long been the slowest corner of the software world. Board bring-up, DeviceTree and Kconfig wrangling, SDK migrations and post-deployment crash forensics are all tedious, documentation-heavy tasks with little tolerance for guesswork. That combination — repetitive work bounded by a large body of written reference material — turns out to be exactly where AI assistants perform best, and vendors have noticed.
Nordic wires AI into the full device lifecycle
Nordic Semiconductor has extended AI-assisted development across its wireless IoT stack, covering the arc from first prototype to deployed fleet. Rather than leaving developers to paste snippets into a general-purpose chatbot, Nordic's approach grounds the assistant in its own material: the nRF Connect SDK documentation, API references, device configurations, and the customer's own field data from nRF Cloud.
The plumbing is an implementation of the Model Context Protocol (MCP). Nordic's MCP servers expose those validated sources to whatever AI assistant a developer already uses — Claude Code, Cursor, GitHub Copilot or another client — so the tooling slots into an existing IDE workflow rather than replacing it. Because the model is working against a specialised, curated corpus rather than trawling generic training data, Nordic says token costs come down as well.
The published use cases are pointedly unglamorous, and that is the point:
- Migrating a project between SDK versions
- Custom board bring-up, including DeviceTree and Kconfig generation
- Diagnosing a crash on a device already in the field
- Identifying faulty devices across a deployed fleet
- Validating release readiness and adding shell commands
Each of these is a task where the answer exists somewhere in documentation or telemetry, and where the engineering time goes into finding it rather than deciding it.
Not a substitute for an engineer
Nordic frames the tooling as assistive rather than autonomous, and early hands-on coverage supports the caveat. In one demonstration of board bring-up, the agent added peripherals nobody asked for — a button and extra LEDs — which had to be corrected by hand. Prompts need to be specific and technical; you cannot vibe-code your way through a hardware configuration.
The practical implication is that these tools compress the time an experienced firmware engineer spends on known-answer work. They do not lower the expertise floor. Reviewing generated code remains mandatory, particularly for anything touching power management, radio configuration or memory layout, where a plausible-looking wrong answer can survive compilation and fail in the field.
A broader shift across the design chain
Nordic is not alone. IoT Analytics expects wider adoption through 2026 of AI-assisted verification, constraint checking and layout optimisation among design teams building edge-AI chipsets, connectivity SoCs and mixed-signal devices. Siemens Digital Industries Software has introduced an AI-enhanced toolset spanning schematic capture to physical implementation, with verification automation, constraint analysis and early flaw detection.
The common thread is scope. The first wave of AI coding tools attacked a single step — generating a function, explaining an error. What is emerging now is assistance stretched across the whole lifecycle, with the same assistant that helped write the firmware also querying fleet telemetry to explain why one deployment cohort is rebooting.
What to watch
Two questions will determine whether this becomes standard practice or stays a demo. The first is cost discipline: several organisations have found agentic tooling more expensive in production than in pilots, which is precisely why vendor-specific, smaller models grounded in curated documentation are an attractive alternative to general-purpose frontier models on every query.
The second is trust in the field-data half of the pitch. Letting an assistant read documentation is low risk. Letting it read and reason over live fleet telemetry raises questions about data residency, customer isolation and what exactly leaves the device. Vendors that answer those clearly will find the enterprise adoption easier.
For now, the honest summary is that AI-assisted tooling is making experienced wireless IoT developers meaningfully faster at the parts of the job they least enjoy — and that this is a genuinely useful place for the technology to land.
Your membership also unlocks:
