It’s the first question everyone asks, and it deserves a straight answer rather than a feature list. The short version: you keep using it. Drafted fixes the part it was never designed to do.
There’s no Drafted chatbot to switch to and no new model to learn. Drafted connects over MCP to the assistant your team already chose, and works the same way in all of them.
Not the quality of the writing, which comes from the same models either way. What differs is everything around it.
| Chat on its own | The same chat, with Drafted |
|---|---|
| Every conversation starts with no idea who you’re | It’s required to search your organisation’s knowledge before it may start |
| Quality depends on how well someone wrote the prompt that day | Your written procedure is loaded first, or the change is refused |
| Output lands in a private transcript only you can see | Output lands on a shared surface your team watches fill in |
| Comparing five options means five scroll positions | Five options sit side by side and are judged in seconds |
| What was it checked against? Nobody can say afterwards | The standard is written down, versioned, and applied before the draft exists |
No future release will add these. They follow from the shape of a conversation itself.
A longer context window is still per-conversation. What your colleague learned this morning can’t reach your session this afternoon, because there’s nowhere shared for it to live.
You can ask a model to follow your procedure. You can’t make it impossible to proceed without one. That takes something outside the conversation, checking before the write lands.
A transcript is single-player. Work that a colleague must check, compare and sign has to exist somewhere they can open without being handed a copy.
The same assistant, working inside a loop that gets better every time you use it.
Before an agent may touch the work it has to search your organisation’s knowledge: your standards, your client history, the decisions you already made.
Your procedure for this kind of work: the checklist, the evidence bar, the house style. Not loaded means the write is refused.
The output lands as frames your team watches appear, compares side by side, and reviews before anyone signs anything.
What you fix goes back into the Wiki and the Skill, so it’s already fixed for every future job and everyone who runs one.
Free to start, and it connects to the tools your team already has open.