Medium launches TK, a writing editor with an @TK AI assistant and transparent AI disclosures

It's a familiar scene: you're writing a draft and need to check something minor, like the number of days in a month. You open another tab, and twenty minutes later you're reading something completely unrelated. Medium has decided to eliminate that detour by building an AI assistant directly into its editor.

How TK works

The name comes from an editorial convention. In journalism, “TK” marks a spot where a fact needs to be added later. It means “to come”, but uses a K because the combination “tk” almost never appears in English words, making the placeholder easy for a copy editor to spot. Now TK is an assistant, too. Type “@TK” in your text, write a request as you would to a colleague in a shared document, and get help without leaving the editor.

According to Medium, it can:

• answer quick questions without pulling you away from your text;
• retrieve passages you previously saved in the app while developing an idea;
• give feedback on your text and headline, and check a draft against the style guides of Medium publications;
• remember and organize everything you've already written, and suggest topics for new pieces.

There's a social layer, too: you can share an early draft with other people and collect their comments. Only the web version is available for now; a desktop app is planned. You can import and export Markdown, but TK itself is not a Markdown editor. The tool is free for Medium account holders, and the company describes itself as profitable.

Transparency instead of bans

The most interesting part isn't the chat assistant inside the editor—those are everywhere these days—but Medium's approach to the debate over authorship. The publishing world is on edge: several authors have lost book deals over suspicions that they wrote with AI, and winners of several literary prizes have faced scrutiny for the same reason. Meanwhile, AI text detectors are unreliable. Models were trained on human writing, and distinguishing between “written with AI assistance”, “written by AI” and “no AI used at all” is very difficult.

Medium takes the view that these distinctions will eventually cease to matter, arguing that AI makes writing more important, not less. At the same time, the company acknowledges the risks: a flood of low-quality content and an increasingly blurred line between human and machine-generated text. Its answer is an author's note that can openly describe how AI was used, down to the prompts and source materials. According to Medium, this is only the beginning of what publishing in the AI era requires: the origins of a text should be clear, transparent and verifiable.

What you can take away

For developers: TK can connect to your own agent through an MCP server or a Claude skill. This lets the app learn about your projects, views and ideas stored in the agent's memory. It's another sign that products are increasingly opening up to users' agents. If you're building an editor, a knowledge base or a CRM, an MCP interface is becoming an expected feature rather than an exotic extra.

For product managers: disclosing AI's role is a practical design pattern. Instead of hiding AI use or banning it, show where it contributed. This reduces the risk of a “did you really write this yourself?” controversy and leaves readers to decide. Medium itself acknowledges that the move could prove contentious, however, and it's still unclear how authors and audiences will respond.

The limitations are worth considering, too. Text in TK is not encrypted because the app is built for collaboration, although access permissions can be configured. If your draft contains trade secrets or unpublished data, think twice. The article does not specify which model powers the app or whether usage limits apply, so it's too early to judge the quality of its suggestions. Finally, an AI disclosure note is an author's self-declaration, not a technical verification: honesty still depends on the person.

Ultimately, TK is not a breakthrough in models. It's a bet on how AI fits into the writing process and how that involvement is explained to readers. For anyone building writing tools, it offers a clear reference point: an assistant that works in context, open interfaces for agents and an honest disclosure of a text's origins.