Introducing Dovetail Coworker

Why does every regulatory question in your company still end up with one person?  Today, Dovetail Coworker is generally available. It brings regulatory help into Slack and Teams, works against your QMS, and keeps our regulatory experts responsible for judgment and review. Here is why we built it.

Someone asks whether a design change needs a new submission. You go and find the documents. You explain the decision you made three weeks ago and the reasoning behind it. You come back with an answer. A fortnight later something changes and you run the whole loop again.

If you are a medical device founder, that is time you are not spending on the product or in front of customers. If you run a larger team, it is your regulatory people reconstructing context instead of applying the judgment you hired them for. Either way the work moves at the speed of one person's availability.

The obvious fix is the one I would be careful with

Point a general AI tool at it. Give it the regulations, give it your documents, let the team ask it directly.

I understand the appeal, and I want to be precise about why it falls short, because the failure is not the one people expect.

Ask a general AI tool whether part of your product is a Class I device and it will give you a clear, confident, well-written answer. The problem is not fluency. The problem is that the answer is built from what is publicly available, and the thing that decides your submission is mostly not public.

Which arguments have been put to a reviewer before. Which ones were accepted, which were rejected, and what the reviewer said when they rejected them. How a particular notified body reads a particular clause this year, which is not always how they read it last year. None of that sits in a published regulation. It sits with people who do this work for a living.

Get classification or scope wrong at the beginning and it does not stay a small error. It propagates into the wrong requirements, verification and validation built against those requirements, and a submission that comes apart under questions you cannot answer.

So a tool that is confident and wrong is worse than no tool, because it removes the moment where someone would have asked.

Headless was the first step

I have written before about headless software, about making our tools accessible to AI rather than only to humans clicking through screens. I still think that matters, and MCP has made it considerably easier.

But connecting your tools to an AI workbench leaves one assumption in place, which is that a person sits in front of that workbench and carries context into it. You still bring the question. You still fetch the document. You still take the answer back to the team.

The tools got connected. The bottleneck did not move.

My thesis, and I will call it a thesis rather than a settled conclusion, is that the next step is not another private workbench. It is putting the expertise into the place where your team already talks to each other. That is a claim about where the working experience ends up, not a claim that MCP is inadequate. MCP can support either shape.

What changes when the knowledge sits in the room

When regulatory help lives in the shared conversation rather than behind one person, a few things change at once.

Your colleagues ask directly instead of queuing behind you. The context is already there, so nobody re-explains the device, the pathway or the decision from three weeks ago. The reasoning stays visible to the whole team rather than disappearing into someone's private chat history. And when an expert weighs in, everyone sees the correction, not just the person who asked.

The founder stops being the messenger. That is the outcome I care about, and it is a different outcome from "the team uses more AI".

What Coworker is, and what it is not

Dovetail Coworker is an agent that works inside Slack and Teams. It joins in when regulatory help is needed, it works against your QMS rather than against the open internet, and it picks up context from how your team actually operates.

The part that matters most is the part that is easy to skip over. Our regulatory experts stay responsible for judgment and review. Coworker is not a replacement for a regulatory lead, and we are not going to pretend it is. It handles the carrying, the drafting and the recall. A human who has sat across the table from a reviewer still owns the call.

That is the distinction between a regulatory coworker and a general AI tool with your documents pasted into it. One of them knows what it does not know and escalates. The other sounds certain.

I will be honest about the limits too. Coworker is strongest where the question touches work already in your quality system. It is least useful on genuinely novel strategy questions where there is no precedent to recall, and those are exactly the conversations where you want an expert in the room rather than a tool.

Coworker is now generally available

Coworker has been running in beta with real teams, and it is now generally available.

If you want to see how Coworker fits your team, you can book a demo. We will show you what it does with your own workflow rather than a demo environment, and we will tell you honestly whether it fits how your team works.

See Coworker in action