We are two engineers. One of us led engineering at a medical device startup that went from its first grant to an ISO 13485 certificate, and saw the whole run from the inside: the company bought the eQMS everyone buys, spent months getting its documents into it, and stopped touching it the week after go-live. The other spent seven years at Google building Fuchsia's networking stack and the conformance tests it had to pass.

Every company that means to sell a medical device has to run a quality system, and almost all of the time it takes turns out to be paperwork, because the tools were built to store the files. Routing an engineer's output into the right SOP and the forms it touches, then chasing two approvals: the quality hires we have talked to spend most of the week on it. At companies too small for a quality hire, it falls to an engineer, who spends weeks writing forty documents from templates they made themselves. Those weeks were supposed to go into the device.

We think a quality system should be the place the work happens, with engineers and the quality lead writing in it together. Writing in it should feel like writing a document, not filling in a form; the system, not the engineer, should know which fields a record needs. The day writing in it is harder than the folder, the folder wins.

And we think the paperwork should be carried by an AI quality engineer that has read every document in your system and the standards you are audited against, and no other company's documents. It takes the engineer's change and drafts it into the SOP and the forms it touches, lines up the approvals, and finds the missing documents and the stale cross-references before the auditor does, so that the people who know the standard spend their hours on judgement instead of receiving forms. It cannot approve anything, and it cannot make a document effective. Some days it will be wrong; saying no to it has to be cheaper than doing the work yourself, and the no is a record too. A person reviews what it did and signs it into document control under their own name. Every draft and finding it makes is its own entry in the audit trail, attributed to the AI, next to the person who accepted or rejected it. The trail gets richer, not thinner. None of this exists yet; this paragraph is the contract we will hold it to.

We are building that. It is called Cove, and the device teams who use it first will shape it. Someone still wears the quality hat; the hat stops being a filing job. If the filing is your week, or you are the one paying for it, we would like to talk to you.

We will ask how your last audit went and what happened the last time a document changed. In return we will tell you, from having done it, which documents you need before your first audit, which can wait, and which of the ones you already wrote will do. If you already have a system and a renewal quote on your desk, we will tell you what moving would actually take and whether you should; if it renews before a design partner has passed an audit on Cove, the answer is not us. If you are a consultant, the call runs the other way: you tell us where your auditor would push, and we show you every action the AI is allowed to take. If you have no system yet and your first audit is still most of a year out, we will ask you the bigger question: whether you want to be the team we build this with; your audit would be the date we build toward. Whichever you are, we will say plainly if Cove is the wrong choice for you.

When there is something to buy, it will be month to month, with a seat for everyone who writes, and your records will leave with you the day you do, as plain files your auditor can read; that holds if we shut down, and if someone buys us, too. Nothing you put in will train anything, ours or anyone's. The contract will say all of it.