ManuscriptRx

A regulated pharma email touches a medical writer, content ops, brand, MLR and CRM. Five teams, five different tools, and two weeks end to end is considered a good run.

This one is mine, not a client's. Nobody briefed it and nothing shipped. I built it because I kept seeing the same AI pitch aimed at this problem and I thought it was aimed at the wrong half of it.

The obvious version is "drop the manuscript in, get an email out." It demos well and it's wrong, because the two weeks aren't spent writing. They're spent waiting at review gates that exist for legal reasons and aren't going anywhere.

So I inverted it. Every step a human is legally accountable for stays with the human. Everything between those steps belongs to the AI: assembling approved content, pulling the market-specific compliance blocks, generating the HTML, running the metadata checks.

That one constraint decided everything else on the screen. It is the difference between a tool that survives compliance review and a demo that doesn't.

The workflow is six steps and every one names who owns it. That sounds like project management rather than design, and in regulated work it is the design. An unowned step is where two weeks goes.

QC sits inline with editing instead of after it. AI auto-pass, Content Ops and Med Writer all sign off while the writer is still in the content, so a rejection arrives while there's still context to act on it.

"Send Preview to Brand" doesn't render until the AI checklist passes. I could have shown it disabled with a tooltip. Hiding it is blunter and it stops the conversation where someone asks to be let through just this once.

Brief Creation is visible in the navigator and locked, labelled "outside pilot scope." The brief is written upstream by people I didn't design for, and drawing a screen for it would have been me inventing a process I hadn't researched.

The PromoMats integration warning stays on screen rather than being tidied away before the screenshot. It's an unresolved dependency, and a reviewer who spots it in a real pilot has a much worse day than one who was told up front.

MLR review itself. I designed the artefacts MLR receives and the state the email is in when it arrives. I did not design the tool they review it in.

How a reviewer annotates a claim, rejects it, and signs their name to that with legal accountability attached is the hardest problem in this whole workflow. It needs access to reviewers I didn't have, and a concept project that claimed to have solved it would be worth less than one that says plainly it didn't.