How to work on several AI drafts at once
Here is the whole idea, before the how: being unfinished with one draft should never block you from starting the next. You work on several documents in parallel — a report, an email, a spec — and none of them holds the others hostage. There's no finishing-tax: you never have to close or complete one to open another.
The chat box doesn't work that way. Every draft lives in its own thread, so working on three at once means three tabs you keep scrolling to tell apart. PassbackAI puts them on one shelf — the bench — and this is how it works.
PassbackAI keeps your recent AI drafts on a bench: one shelf you switch across from the top bar, holding the last ~20, most-recent first, grouped into Pinned / Today / Yesterday / This week / Earlier. Different documents or successive versions of one — both live here, and being unfinished with any of them blocks nothing. Pin the one you can't lose and it floats to the top and can't roll off. Paste a new version of a draft already on the bench and the Auto-Diff auto-detects the match and paints exactly what the AI changed. It all stays in your browser. Paste an answer to start a bench.
The morning of three buried threads
Picture a normal Tuesday. You have three things in the air, and the AI touched all of them. A client proposal you drafted with Claude last night. A product spec you and ChatGPT hammered out this morning. A tricky reply to a stakeholder you asked the model to soften. Three real pieces of work, all mid-flight.
Now you need the spec — and you're scrolling. Was it the thread from 9am, or the one where you also asked about pricing? You open two, read the first lines of each, guess wrong, go back. The chat UI files every draft behind its own scroll, in its own conversation, with no shelf where they sit side by side. Each one is somewhere; none of them is here. The tax isn't writing the drafts — it's finding which thread the right one is buried in.
That hunt is what the bench removes. It's one place your recent drafts sit together, so switching between the proposal, the spec, and the reply is a click, not an excavation.
What the bench is
The bench is your recent drafts on a single shelf. Paste an AI answer and it lands there; paste another and it joins it. You move between them from the document switcher in the top bar — open it and the whole shelf is there, each draft titled from its own first heading, ready to reopen exactly where you left it, every highlight and note intact.
You don't even have to open it. Once two or more drafts are on the bench, ↑ jumps to the previous draft and ↓ to the next, opening it the instant you press it — ⌘[ and ⌘] mirror them if that's the back/forward muscle memory you have. You flick between the proposal, the spec, and the reply without lifting your hands from the keyboard, and without closing anything first. That's the north star made literal: the next draft is one keystroke away, not a save-and-close ritual. (Press ? anytime for the full list of shortcuts.)
The switcher groups the shelf so a long bench stays scannable — not one flat list, but five buckets by recency:
- Pinnedthe drafts you chose to keep with you, always at the top
- Todayeverything you've touched since local midnight
- Yesterdaythe single prior day
- This weekthe recent tail behind that
- Earliereverything older still on the shelf
The shelf holds the last ~20 drafts. That's a deliberate ceiling, and it's what makes the next part — pinning — matter.
Unfinished isn't blocked
This is the point the whole feature is built around, so it's worth saying plainly: you never have to be done with one draft to start the next. In a chat UI, moving on has a subtle cost — the draft you were on recedes up the thread, and starting fresh somewhere else means one more conversation to keep track of. So you finish things you'd rather have parked, or you park things and lose them.
On the bench there's no such trade. Leave the proposal half-marked, open the spec, jump to the reply, come back — the proposal is right where you left it, in the same state, one click away. Parallel work is the default, not a workaround. And "the next draft" cuts two ways, both of them first-class:
- Different documents. The proposal, the spec, and the reply are three unrelated things, and they coexist happily. The bench doesn't care that they're different — it just keeps them within reach.
- Versions of the same document. You ask the model for another pass at the spec and paste the result. Now v1 and v2 both sit on the bench. Neither replaces the other, and — as the next section shows — the tool notices they're kin.
Pin the one you can't lose
The shelf is bounded at ~20, and it self-cleans: when a new draft arrives past that ceiling, the bench drops the least-recently-touched one to make room. Crucially, it never drops the draft you're actively on, and it never touches a pinned one — the oldest, most-ignored draft is what rolls off. For everyday clutter that's exactly right; the throwaway you pasted last week shouldn't outlive its usefulness.
But some drafts matter beyond their recency. You have a meeting on that proposal next Thursday, and between now and then you'll touch a dozen other things that would otherwise push it down and eventually off the shelf. That's what pin is for. Pin a draft and two things happen:
- It floats to the top, into the Pinned bucket, so it's the first thing in the switcher no matter how much else you paste.
- It's exempt from eviction. The bench skips pinned drafts entirely when it needs room, so the one you flagged for the meeting is still there when the meeting comes.
One detail that makes pinning feel right: pinning doesn't count as touching the draft. It lifts the draft to the top of the shelf without pretending you just edited it, so its place in the Today / Yesterday / Earlier timeline stays honest. Pin is "keep this with me," not "mark this as fresh."
See exactly what the AI changed between two versions
Here's the payoff that ties the bench to that second case — versions of the same document. When you paste a draft that closely resembles one already on the shelf, PassbackAI notices on its own. It compares the new paste against what's already on the bench, and if one is clearly a revision of another, it offers the Auto-Diff: it lines the two versions up and paints exactly what changed — what the model added, cut, and rewrote — so you can step through the edits one at a time.
This is genuinely hard to get any other way. When a model hands you "here's the revised spec," it rarely tells you the truth about what it touched — it says it tightened the intro and you have to take its word, or re-read all 1,800 words to check. The Auto-Diff is the only convenient, visual way to see what actually moved between two versions, instead of trusting the model's summary of its own edits. Paste the new version, and the changes light up.
The detection is deliberately conservative — it would rather stay quiet than line up two drafts that aren't really related, so it only offers the diff when the match is strong. And a related nicety falls out of the same machinery: if you paste the exact same draft again, the bench reopens the one you already have instead of spawning a duplicate. The shelf stays the drafts you're working on, not a pile of accidental copies.
What the bench is not
The bench is a working shelf, not a filing system, and it's more useful once you know where its edges are:
- It's not cloud storage or a document manager. There are no folders, no archive, no search-your-whole-history. It holds the drafts in front of you right now, not everything you've ever pasted.
- It's not permanent. The shelf is the last ~20, and the oldest untouched draft rolls off as new ones arrive. If a draft needs to outlast that, pin it — or better, get its feedback back to the model and let it go. Drafts here are meant to be in transit.
- It's not synced across your devices. The bench lives in the one browser you're using; open PassbackAI on your phone and you won't find your laptop's shelf there. It's local, which is also why nothing on it is uploaded.
When a draft needs to leave the bench — to reach a colleague, or to go back to the model as one anchored batch of notes — those are their own deliberate steps (sharing, routing, copy-back), not something the shelf does silently in the background.
Where this fits
The bench isn't a feature you go turn on. It's just what happens the second time you paste: the first answer stays, the new one joins it, and you have a shelf. From there the moves are small — switch from the top bar, pin what you can't lose, paste a new version to diff it — and they add up to working the way you actually work, on several things at once, without the chat box's one-thread-at-a-time tax.
Try it: paste an answer, then paste a second one, and open the switcher in the top bar. You have a bench.
Work on several drafts in parallel. Not being done with one never blocks the next — that's the whole point of the bench.