Trust, but verify.

Since 2021, we have built software for regulated industries, where every conversation with a customer or prospect has to hold up to the rules. Our job was compliance: knowing what licensed agents said, when they said it, and whether it would pass an audit.

In late 2024 we put AI voice agents on the front line to take the most repetitive and demanding calls, then hand off to licensed agents. That is when the real lesson landed. Launching a voice agent was the easy part. The hard part was proving it stayed compliant on every call, including after a prompt change, a new version of the underlying model, or a swap in the speech-to-text, voice engine or text-to-speech pipeline. Any one of those can quietly change how an agent behaves.

So we built a supervision layer for ourselves. It checked every AI conversation against the rules of the industry, flagged a gap the moment it appeared, and gave us what we needed to act. We called it Supervisr.

In early 2026, Loïc joined to make sure that what we had built for engineers would be just as clear to the compliance and operations people who live in it every day.

Today, Supervisr does for other teams what it did for ours. You can trust your AI voice agents because you can verify them. Every call. Every day.