Medium Opens TK to All Members With Inline AI Assistance

By Sophie Lindqvist

Medium announced on November 9, 2026 that TK, its private notebook and writing app, is free with a Medium account after an invite-only phase. Medium’s release puts AI research, critique and retrieval inside the private work that usually comes before publication. For Medium, that extends its role from hosting published work into the private process that precedes it.

What changed on November 9?

Medium’s September 16, 2026 post described TK as invite-only or available to Friends of Medium while the company prepared broader access. The November announcement ends that staged rollout for anyone with a Medium account.

TK remains separate from Medium’s publishing platform. Chief executive Tony Stubblebine describes the division this way: “TK is the gym and Medium is the stage.”

Writers can practice ideas in the notebook before deciding to publish them on the larger platform. The practical change is wider access to a separate product.

Why does trust shape the product?

Medium’s State of Writing Report 2026 found a sharp gap between adoption and trust. In its Harris Poll survey, 71% of surveyed writers had used AI as part of their writing during the previous three months.

Yet 60% distrusted their own AI-produced writing, and 69% distrusted such writing from other people. Medium’s design pairs private notes with controlled sharing and provenance records to address that distrust.

The Harris Poll conducted the online US survey for Medium from August 12–17, 2026, among 1,175 writers. Medium reports full-sample precision of 4.0 percentage points, using a 95% Bayesian credible interval.

The report also counted 4.4 million unpublished drafts written on Medium over the preceding year. That volume gives TK a clear role: hold private work, let AI support it, and leave publication as a separate choice. Provenance records address the survey’s distrust.

Medium’s product design answers the survey’s tension. Writers already use AI, but most of those surveyed did not fully trust what it produced.

What ships in the first version?

Medium says the first version includes:

• a mobile app and widgets for voice and text capture;
• imports of Medium archives and Bear or Obsidian vaults;
• Markdown import and export.

It ships without a native desktop app. Stubblebine said Medium wants to add one. TK’s first version supports mobile capture and Markdown files, while desktop support remains a goal.

How does @TK work inside a draft?

Writers can type @TK in the middle of a sentence to check a fact or retrieve earlier writing without leaving the editor. Medium says the name adapts the publishing placeholder TK, meaning “to come,” which writers traditionally leave for unfinished information.

The same placeholder can now trigger the assistant. Medium says this keeps research and retrieval from interrupting the draft.

AI feedback covers editorial and headline suggestions. It can compare drafts with Medium publications’ style guides, and it can flag material that could be organized into a book.

Medium says TK will not write a user’s essay. The company positions the assistant as support for the writer’s own work rather than a substitute for it.

The assistant checks, suggests and retrieves. The essay stays with its writer.

What can TK do with an archive?

TK creates a writing prompt each morning, and the prompt can refer back to earlier notes. Searches, topics, tags and libraries can become wiki-style summaries with themes, key ideas and links to the underlying notes.

Medium says those summaries give an unstructured archive a route back into later drafts. TK turns saved material into a resource for new writing, rather than a pile of disconnected records.

That changes the role of a notes archive. TK gives it a way to feed the next draft.

Can personal agents work inside TK?

Medium says TK includes an MCP server and a Claude skill. The connections let personal AI agents draw on a user’s stored positions, projects and unfinished ideas.

Medium’s Claude Daily Note plugin shows how one connection works. With TK’s MCP server connected in Claude Cowork or Claude Code, the plugin creates one document per day and updates it in place.

The plugin also carries unchecked to-do items forward. Medium says its targeted edit operations preserve live edits, existing checkbox states and the user’s writing.

TK connects outside agents to stored ideas while its edit tools are designed to protect work in progress. That makes the notebook a bridge to tools that work with a writer’s own material.

Who controls sharing and AI training?

Social libraries let writers share drafts with groups. Medium says these include open libraries for public journaling, feedback before publication and developing a self-published book.

Notes otherwise remain private, with access permissions under the writer’s control. Medium states that no AI company trains on or uses a person’s writing without permission.

Published TK notes can disclose which material was human-written and which was AI-written. The provenance record goes deeper: writers can inspect the prompts and source material behind the note.

That level of detail gives readers a way to judge the note’s authorship. It also gives writers a way to explain how a tool contributed. Sharing remains deliberate, and published notes can carry a record of human and AI authorship.

Why does Medium need a private app?

Medium already hosts finished work. TK extends the company into the private process that comes before publication, where ideas remain rough and unfinished.

Stubblebine’s gym-and-stage shorthand captures that division. AI can check facts, retrieve old notes, summarize an archive and critique a draft, while the writer keeps authorship.

TK then adds a second layer of control. Writers decide which notes become shared libraries, which published notes disclose AI involvement, and who can access their stored work.

The stakes sit in the writer’s workflow, where AI can shape a draft before publication. Medium still wants a native desktop app. The bigger question is whether writers will trust AI with unfinished work.