Medium’s new TK app brings AI assistance directly into the writing process

Online publishing platform Medium launched a dedicated writing application named TK to integrate artificial intelligence directly into drafting and drafting-stage research, as detailed in TechCrunch's report. The product embeds an assistant into the text editor, allowing writers to trigger inline interventions by typing @TK followed by instructions or questions. The company named the software after the traditional journalism placeholder denoting information that is to come, positioning the software to help authors systematically address gaps in working manuscripts.

This report examines how the new editor functions, its workflow capabilities, its connection to external agents, and its release specifications for writers and digital publishers. The tool targets long-form writers and Medium contributors seeking draft-level research, contextual notes, and style-aligned editorial feedback. By combining machine assistance with disclosure tools, the software addresses increasing industry tensions surrounding manuscript provenance and human authorship.

01Editor prompts and placeholder lookup

The core interaction in TK relies on typing @TK directly in the composition window to invoke the assistant without leaving the manuscript. Writers can query simple factual information that typically requires opening a separate browser tab, such as checking the number of days in a particular month. The tool also indexes notes and fragments previously stored in the workspace, retrieving relevant background material while authors flesh out active arguments.

Beyond quick queries, the assistant incorporates specialized skills tailored to publication standards. The software provides headline feedback, editorial critique, and evaluations that align an active draft with specific style guides used across Medium publications. These built-in assessments evaluate drafts during composition, functioning alongside basic proofreading utilities.

Writers can also request writing prompts or use the software to organize past material. By retrieving prior notes, the editor keeps research and narrative arcs in view, reducing friction between conceptual note-taking and final execution.

02External agents and memory connections

The software extends past standalone generation by offering optional links to external AI agents. Users can connect personal agents through TK's Model Context Protocol server or via a dedicated Claude skill. These connections let external models feed their accumulated memory and custom instructions into the writing environment.

By integrating through the Model Context Protocol, the assistant references an author's distinct projects, specific positions, and saved background knowledge. This architecture grounds inline responses in the author's own reference materials rather than relying only on generic model data.

Medium pairs this agent layer with tools that retain past drafting activity. Authors can review prior iterations and maintain consistency across complex subjects without manually re-entering contextual boundaries for each new piece.

03Social feedback and draft provenance

The application includes a social layer designed to collect human reactions alongside automated checks. Authors can distribute early drafts to colleagues and collaborators to gather critiques before publication. This structure positions automated editorial feedback alongside input from human readers.

To address transparency concerns regarding machine-assisted drafting, TK introduces notes that detail how AI was deployed across a document. Authors can publish inspectable records detailing prompts and original source material, documenting how a final piece came together.

Medium acknowledged concerns that generative tools risk flooding platforms with low-quality copy and blurring lines between human and machine prose. The provenance system aims to make the underlying workflow auditable right down to initial prompts and reference texts.

04Publishing debate and detection limits

The release arrives amid industry-wide scrutiny over artificial intelligence in book deals and literary competitions, where suspected machine involvement has led to revoked awards and cancelled contracts. Publishers increasingly worry that automated editorial recommendations can displace the author's voice, turning suggestions into the text itself.

Complicating the debate is the recognized unreliability of AI detection software, which struggles to separate machine-assisted human prose from purely automated generations. Because models train on human text, automated detectors frequently deliver conflicting or inconclusive assessments on edited manuscripts.

Medium argued that parsing these minor distinctions will ultimately lose relevance as assistive tooling normalizes. The platform maintained that artificial intelligence elevates the importance of the written word and declared that the technology amplifies human intelligence through written language.

05Web rollout and technical limits

TK is available immediately on the web as a free tool for users holding an active Medium account. Medium confirmed that a dedicated desktop application is currently on its development roadmap, expanding access beyond the browser interface.

The application supports importing and exporting files in Markdown format, though Medium clarifies that the editor is not an entirely native Markdown tool. Documents created in the editor are not encrypted, reflecting its focus on real-time sharing, though creators can configure specific access permissions for their documents.

The profitable publishing platform is deploying the tool without additional subscription charges for account holders. The rollout combines document access controls with collaborative sharing, allowing authors to retain permission management across web-based project drafts.