Patterns, scenarios, and lessons on developing connected products with AI agents, validated on real hardware.
Vibe coding is a way to find out whether an idea is worth building. Agentic engineering is how a product that has to keep working gets built: plan, review, implement, review, with documentation and validation on the real product in the loop.
A practical guide for engineering leaders getting their team started with AI coding agents on a connected product. Before the first task, have the agents write and maintain a Markdown ground truth of the system; then hand them a mechanical, pattern-heavy task, set it up with the right examples and constraints, and put in-loop validation around them so they check their own work and get far more right on their own.
A connected product team already tests at several levels: unit tests on each module, full-system scenarios on real hardware, and a QA pass before release. An in-loop validator has to reach across all of them, so the setup that keeps AI agents fast runs every one of those tests as the work lands, not just at release.
The scenarios that verify a connected product run from inside the loop where each change is made. AI agents run them on real hardware and read the results as they work, and that placement is where the reliability, the speed, and the confidence come from.
Three green checkmarks from three test suites don't add up to product-level confidence. The gap is between the domains, outside the ground any one test suite covers.
AI agents write a firmware change in an hour that used to take a week. Then it sits three days in a manual cross-domain verification queue. A team ships at the speed of its slowest verification step, and for a connected product that step is cross-domain verification.
When code output gets cheap, the scarce resource is the judgment around it: what to build, what quality bar to hold, how to verify the output. Those are leadership decisions, and they run on institutional knowledge no agent has.
An AI agent that runs the cross-domain scenarios while it works hands over changes already validated on real hardware. Code now arrives at software-company speed wherever teams put agents to work. The release process decides whether the product ships at that speed too.