An agent that joined a regulated software team

A European software company working under medical-device regulation carries a load that has little to do with writing software. Its procedures live in a quality system and have to match what people actually do. Adverse events have to be recorded and assessed. And while clinical trials are live, the infrastructure behind them raises questions that someone has to answer.

The company brought Otto into its team channels. It did not arrive as a new tool with its own login. It arrived in the chat the team already used, where people could ask it things the way they ask a colleague. That is how our method starts: an agent that learns from and with a team, asks questions, builds a model of how things work, and offers to take on work only where it can add value.

Within weeks, colleagues were bringing it real work. A few examples, as they happened.

It audited the company's clinical procedures and posted its findings to the quality channel, tagging the right people. The reply was short: "Thanks Otto, good work!"

It went through adverse-event exports row by row, checking each entry. It flagged only two real gaps. One colleague wrote: "great that there was clearly only 2 that Otto tagged, we have done good job with our risks." Another added that it is good to go through manually too, "but I've found that Otto is quite good with these."

It explained how to set up the office VPN, working from the company's own internal document rather than from general knowledge. "this was very helpful, thanks Otto."

When a cloud provider made an announcement, it explained what that meant for the company's credits. "thanks Otto, that was very helpful, will discuss this tmrw."

And because a team is not only its procedures, it once made a meme and delivered it to a colleague end to end. The verdict: "im impressed."

What made this work in a regulated setting is that Otto reports rather than decides. It writes in the team's internal channels. When it finds a problem in a procedure, it tells the people who own that procedure rather than changing it. When it checks a table, its findings go to a person, who can still go through them by hand. Its working rules are kept in files, so when it tightened one of them, the change was there for the team to see. A colleague's first question after seeing it do that was: "Ok wow. Where is all this running?"

If your business carries this kind of load, click here to let us know more about your business.