That's the idea behind how aihops handles AI, and we think it's going to matter more as AI writes more of what businesses send.
Every AI tool wants to be the whole workflow
Each AI tool is building its own way to share what it makes. That's useful, up to a point. But follow it through and the document lives inside one vendor: the person reading it needs an account there, the comments stay there, and when you ask for a review, the model checking the work is the one that wrote it.
That's fine for a draft you're keeping to yourself. It isn't how a proposal reaches a customer, a contract gets past legal, or a board paper gets approved. Those involve people in other organisations, using other tools, or none.
Three ways to be agnostic
About where a document came from
Claude, ChatGPT, Gemini, Copilot, a script of your own, or a person with a text editor: aihops doesn't mind. Upload HTML, PDF or Word, paste it, or email it in. Claude and ChatGPT can connect to aihops directly, as equals, and neither is required.
About who reviews it
The people you share with need an email address, and that's all. No AI account and no subscription, and they don't have to use AI at all to take part. A customer's general counsel marks up your proposal the same way your own team does.
About the lens it's read through
An aihops AI review is done as a persona: who they are, what they're responsible for, and the standards they hold documents to, written down once. "Group counsel, who holds contracts to our limitation-of-liability position." "A CFO who checks every number adds up." "A sceptical customer who's been burned before." The persona is yours. It doesn't belong to the tool that drafted the document, and it reads every document the same way, wherever it came from.
A second pair of eyes shouldn't be the first pair.
Why the reviewer shouldn't be the writer
Ask the conversation that wrote a document whether it's any good, and it's marking its own homework. It knows what it meant to say, so it reads that, not what's on the page. It shares its own assumptions, and it remembers the shortcuts it took and why they seemed fine.
An aihops review starts cold. It sees the document as reviewers do, with none of the drafting conversation, and a stated point of view to read it from. To be plain about it: aihops's own reviews use Anthropic's Claude today, so a document drafted in Claude is reviewed by the same family of model, with a fresh context and a different brief. What makes the review independent is that it's separate from the writing, done from a point of view you set, and checked by people. It isn't a claim about which model is better.
People and AI on the same page
AI findings sit beside everyone else's mark-up, pinned to the words they're about. They're labelled AI and say who asked for them, "from Claude as CFO, requested by Sam", so nobody mistakes them for Sam's own review. People reply to them, argue with them and resolve them like any other thread.
When an owner asks AI to consolidate everyone's mark-up into the next version, it shows its working: each proposed change says what it replaces, why, and which comments it answers. The owners keep or decline each one. Nothing changes until a person says so.
The record outlasts the tool
Tools change. The model you use this year won't be the one you use in two, and your customer's company may standardise on something else entirely. On aihops, the record of how a document came together, every version, every comment, every proposed change and why it was kept or declined, lives with the document, not with the tool that wrote it.
And it's yours to take anywhere. Export for Claude gives you the comments, history and current version as JSON and markdown that any AI tool reads, and the aihops connector works from Claude and ChatGPT alike.
What this means in practice
- Draft with whichever AI suits the job. Switch when something better comes along.
- Share with anyone, whatever they use, without asking them to sign up for your AI tool.
- Write your standards down once, as personas, and hold every document to them.
- Treat AI's findings as one voice in the review, labelled and checkable, not the verdict.
- Keep the history of each document with the document, where the next tool can read it.
AI makes the drafts. People decide what goes out. aihops sits between the two, whichever tools they use.