Medium TK puts AI in the editor. The real product is provenance

In newsrooms, “TK” means “to come”, the placeholder for the number you'll fill in later. Medium named its new writing app exactly that. And it ships with an answer to a question every brand has right now: who actually wrote this?

What's new

• The app: Medium launched TK on November 9, a writing app with an AI assistant in the editor. You call it with “@TK” plus a request.
• What the assistant does: answers quick research questions, pulls in writing you saved earlier, gives editorial and headline feedback, and checks a draft against the style guide of Medium publications.
• Open to your agents: if you want, you can connect your own AI agent through TK's MCP server or a Claude skill.
• Provenance: authors can publish notes on how they used AI. Medium wants a piece's provenance to be “clear, transparent, and inspectable”.
• Availability: web at launch, desktop app on the roadmap. Free for anyone with a Medium account.

The provenance note is the feature, not the assistant

Sounds like the hundredth AI writing tool. Not quite. An assistant that critiques headlines is everywhere. What's rare is a platform that makes disclosure part of the writing process instead of hiding it in a footnote.

For me, trust is a brand's most valuable asset. AI content belongs where disclosure and quality are right. That's why every article on i4u.business carries an AI note, and AI images are labelled. Medium is turning that stance into a feature. And if you disclose, nobody can catch you out.

The style guide becomes the operating system

Here's what I find interesting: the assistant checks against the publication's style guide. A model only knows what you hand it. If your style guide is a PDF nobody has read, no AI can check against it. Teams with clear rules on tone, terms and no-gos get an editor that never gets tired. Teams without rules just get mediocre copy faster.

Medium itself says AI makes writing more important, not less. At the same time, it concedes that AI could mean more low-quality content and blur the line between human and AI writing. You can't state the dilemma much more honestly.

The catch

Writing in TK isn't encrypted, because the app is built for collaboration; access permissions can be set. For confidential corporate drafts, I'd clarify that internally first. And a provenance note is only as good as the honesty of whoever writes it.

What I'd do now

• Write an AI note for your content. Two sentences: what you use AI for, who checks and approves. Under every blog post, or on its own page.
• Make your style guide machine-readable. Tone, terms, banned words, three good example texts. Short enough to fit into every prompt.
• Set a rule for drafts. What can go into external writing tools, and what stays internal?

Bottom line: The question is no longer whether you write with AI, but whether you say so before someone else does.