The Thinking Room Companion

The Thinking Room Companion is a self-reported ChatGPT app built for use within OpenAI's platform during Build Week 2026. The author describes it as an application that structures AI-generated responses into visible, accountable judgements by separating conversation from commitment. It uses GPT-5.6 to propose structured entries (e.g., evidence, interpretations, challenges) while requiring human confirmation before anything enters the accepted record. The system enforces a "human-in-the-loop" boundary where AI can interpret and propose but not accept its own proposals.

The description states that this is a prototype built during a hackathon event with no independent verification or traction data. It is positioned as an educational tool aiming to make reasoning visible in AI-assisted decision-making, particularly for students evaluating research claims or professionals making judgments based on information.

The single most important open question is whether the described interaction model — separating AI interpretation from human acceptance — actually works in practice when used by people who did not design it. There is no evidence of user testing beyond the author's own iteration process.

The description states that the Thinking Room Companion is a ChatGPT app built using OpenAI Apps SDK, Model Context Protocol (MCP), TypeScript, React, Prisma, PostgreSQL, Zod, Render, Vitest and Testing Library. It operates as an embedded widget within ChatGPT.

The system allows users to contribute naturally in ChatGPT, where GPT-5.6 acts as a "Chief" interpreting that contribution provisionally and proposing structured entries for a visible Thinking Room. The human, acting as Chair, reviews those proposals before anything becomes part of the accepted record.

Proposals are organized into categories including:

• Evidence
• Interpretations
• Challenges
• Uncertainties
• Judgement
• Ownership

The result is described as a "human-owned Judgement Record" that shows what evidence and reasoning the human was prepared to accept, what remains uncertain, and who owns the decision.

The description states that the product is positioned as a tool for making human judgement visible in AI-assisted environments. It distinguishes itself from conventional AI assistants by conducting structured inquiry rather than producing immediate responses.

Key claims include:

• "Abundant answers increase rather than remove the need for human judgement"
• "The Thinking Room is a human-governed judgement system inside ChatGPT"
• "Most AI assistants present an immediate response. The Thinking Room conducts an inquiry and separates conversation from commitment"

The positioning evolved from a conceptual framework described in the author's book series "Thinking in the Age of Answers" to a practical implementation during Build Week.

The description states that the target use case is for moments when an answer begins to carry weight, particularly in education where students evaluate research claims. It also mentions potential applications in research supervision and product review.

The author notes that the current prototype demonstrates the interaction model but does not yet claim proven educational effectiveness. The system appears designed for users who need to make decisions based on information and want to make their reasoning process visible.

Not evidenced. The description provides no information about pricing, monetization or business model.

The description states that the application was built using:

• TypeScript and React
• OpenAI Apps SDK and Model Context Protocol (MCP)
• Prisma and PostgreSQL for persistent room state
• Zod for typed input/output validation
• Render for deployment
• Vitest and Testing Library for testing

It includes features such as:

• An embedded widget in ChatGPT
• Persistent PostgreSQL-backed room state
• Complete proposal lifecycle with accept, amend, reclassify, remove and reject actions
• Enforced human-confirmation boundary
• Separate Judgement and Ownership records
• Bidirectional consultation grounded in accepted state
• Markdown and PNG exports from accepted room state

The implementation involved a repeated loop of describing intended behavior, building it, encountering what actually happened, and deciding what needed to change.

Not evidenced. The description states that this is a prototype built during Build Week with no independent verification or traction data beyond the author's own iteration process. It explicitly notes that "Those questions cannot be answered by adding more claims to the project description. They need to be answered through use."

The description states that most AI assistants present immediate responses, while the Thinking Room conducts a structured inquiry and separates conversation from commitment. It positions itself as different from conventional AI assistants in how it handles reasoning and commitment.

No specific competitors are mentioned or described in the text.

• The system is described as a prototype built during a hackathon with no independent verification
• There is no evidence of user testing beyond the author's own iteration process
• The description states that "The first version worked technically, but it did not feel like a room" indicating potential usability issues
• The author explicitly notes that "Those questions cannot be answered by adding more claims to the project description. They need to be answered through use"
• The system requires human confirmation for all accepted material, which may slow down workflows or create friction

1. What specific user testing has been conducted beyond the author's own iteration process?
2. How does the system handle edge cases where AI proposals are ambiguous or contradictory?
3. What is the actual implementation of the "human-in-the-loop" boundary? How is it enforced in code?
4. Has there been any evaluation of whether users can actually distinguish between conversation, proposal and accepted record?
5. What are the specific challenges encountered during the development process that were not mentioned in the description?
6. How does the system handle situations where the human rejects a proposal but then wants to revisit it later?