Stet has been introduced—a new Markdown editor for macOS that changes the approach to human-AI collaboration. Instead of automatically overwriting text, the tool uses a system of proposals that requires explicit user confirmation for every edit.
What Happened
Developers have released Stet, which focuses on the safe use of generative AI when editing documents. The application works with local files and supports integration with any AI agents, including Claude, ChatGPT, and local models, without the need to use specific APIs. Instead of directly changing the text, Stet generates proposals at the paragraph level, which the user must either accept or reject.
Context
In modern workflows, using LLMs often leads to a loss of control over authorship and unpredictable replacement of original text. Stet implements a human-in-the-loop pattern, acting as a UI layer that abstracts interaction with various models and shifts the focus from full automation to controlled co-authorship.
Why It Matters for the Industry
For the industry, this represents a significant shift in generative AI UX: moving from a "generation on behalf of the user" model to an "assisted with confirmation" model. The product demonstrates a viable interaction pattern that could become a standard for professional tools where accuracy and preservation of authorial style are critical.
Why It Matters for Users
For users working with Markdown (article authors, developers, researchers), Stet allows for the integration of AI into a familiar local workflow while maintaining full control over every edit, without trusting third-party agents to automatically modify their notes or code.
What Is Not Yet Known / Limitations
From an architectural standpoint, the solution is a UI layer and does not offer new algorithmic or architectural innovations in the field of machine learning.
Sources
Author
Look at AI, Editorial Team