AI1AI-Assisted Embedded Development
Master AI coding agents for embedded systems
Objectives
|
- C language
- Embedded target experience
- Git basics
- Command line
- No AI experience needed
- Theoretical course
- PDF material in English (printed for face-to-face); online over Teams.
- Trainer assistance throughout.
- Practical activities (40-50% of duration)
- Code examples, exercises and solutions.
- Remote: one online Linux PC per trainee, with emulated or physical board as needed.
- Face-to-face / on-site: one PC (one per two beyond six trainees), target board and install manual as needed.
- Downloadable preconfigured VM to redo the labs afterwards.
- Each session starts with a trainee check-in.
- Any embedded systems engineer or technician with the above prerequisites.
- Prerequisites are checked before the training.
- Progress is assessed by the trainer through the practical exercises, and by quizzes for sections without exercises.
- Each trainee receives a completion certificate.
- If a prerequisite gap appears, alternative or additional training is offered.
Course Outline
- AI assistant families
- Interaction modes
- Why embedded is different
- From coding to orchestrating
- Strengths and limitations
| Exercise: | give the same driver prompt to AI tools and compare the results | |
- Anatomy of a prompt
- Few-shot prompting
- Iterative refinement
- Prompting with embedded artefacts
- Reviewer-mode prompting
- Plan before code
- Detecting hallucinations
| Exercise: | rewrite three requests as structured prompts and measure the improvement | |
- Why specs come first
- Anatomy of a spec
- From requirement to spec
- Spec as contract
- Living specifications
- Role of the FSD
- FSD structure
- Generating a first-draft FSD
- Reviewing the FSD
- Iterating the FSD
- Handing off the FSD
| Exercise: | turn a feature request into a structured specification | |
| Exercise: | find the hidden assumptions in a requirement and make them explicit | |
| Exercise: | generate a first-draft FSD with Claude | |
| Exercise: | review and harden the FSD into a ready-to-build version | |
- GitHub Copilot in VS Code
- Claude in VS Code
- Configuring the IDE
- Privacy and licensing
- What a CLI assistant adds
- Scaffolding a new firmware project
- Claude Code
- OpenAI Codex CLI
- Project memory files
- Encoding the constraints
- Permissions and autonomy
- Slash and custom commands
- Hooks (build, clang-format, lint)
- Plan Mode and Extended Thinking
- Claude Code on the web
- Context-window economics
| Exercise: | set up Copilot and Claude in VS Code, then build an I²C driver with each and compare | |
| Exercise: | configure CLAUDE.md and test hooks, then let Claude Code build a small driver | |
- The datasheet problem
- Feeding the right pages
- Per-project knowledge corpus
- Trust heuristics
| Exercise: | generate a DMA config | |
- MCP as the agent’s interface
- MCP server categories
- Designing the MCP toolbox
- Writing an MCP server
- Sharing MCP configs
| Exercise: | connect Claude Code to three MCP servers and verify each one | |
| Exercise: | write a small MCP server | |
- What an agent is
- The agent loop
- Single vs multi-agent
- Subagents and delegation
- Autonomy boundary
- Agent configurations
- Failure modes
| Exercise: | build a two-agent workflow (one implements the FSD, the other reviews it) | |
| Exercise: | run an agent on a build-flash-test loop, inject a failure, and watch it recover | |
- The Skill concept
- Skills for embedded teams
- Installing and using plugins
- Plugins vs Skills
- Project-level config files
- Encoding team standards
- Versioning and sharing Skills
- Three tools, one workflow
- Where each fits
- Switching tools mid-task
- Decision guide
- Cost and licensing trade-offs
- IP and data residency
| Exercise: | write a Skill that turns a one-line request into an FSD, and test it on three cases | |
| Exercise: | write a CLAUDE.md of coding conventions and see how Claude Code’s output changes | |
| Exercise: | take one feature through all three tools (Claude for the FSD, Codex to build, Copilot to refactor) | |
- What review means
- Hardware failure modes
- Concurrency failure modes
- Timing failure modes
- Review workflow
| Exercise: | find and fix three planted defects (register, concurrency and timing) in a generated driver | |
| Exercise: | have a second AI tool review the first’s output and compare what each missed | |
- IP and data handling
- Your company’s AI policy
| Exercise: | classify ten scenarios as safe, borderline or forbidden, and draft a one-page house rule | |
More
To book a training session or for more information, please contact us on info@ac6-training.com.
Registrations are accepted till one week before the start date for scheduled classes. For late registrations, please consult us.
You can also fill and send us the registration form
This course can be provided either remotely, in our Paris training center or worldwide on your premises.
Scheduled classes are confirmed as soon as there is two confirmed bookings. Bookings are accepted until 1 week before the course start.
Last update of course schedule: 27 June 2026
Booking one of our trainings is subject to our General Terms of Sales