Whetstone

What the company appears to be: Whetstone is a self-reported AI-powered thinking partner tool built for personal idea development and introspection. The author describes it as an application that interviews users in a structured way, using GPT-5.6 to challenge assumptions without flattery, citing user-generated content back at them. It includes features like a "Mirror" (a reflective letter based on conversation) and a "Workshop" (structured idea critique), with a focus on anti-sycophancy as a core engineering principle.

What changed: The project was rebuilt from scratch using Codex for code generation, with an emphasis on spec-driven development, testable prompts, and deterministic behavior. It moved away from no-code approaches to a more rigorous engineering process involving guardrails, eval suites, and deployment pipelines.

The single most important open question: Is there evidence of real user adoption or traction beyond the author’s own use cases? The description does not include any data on users, revenue, or customer engagement — only claims about product functionality and internal testing.

• The description states that Whetstone is an AI tool designed to act as a thinking partner.
• It conducts discovery interviews one question at a time, beginning with a mood check shaped by the user's own words.
• It builds a "Mirror" — a letter summarizing where the user’s energy spiked or flatlined, quoting the user back and surfacing unnamed patterns.
• It includes a "Workshop" feature for refining raw ideas through structured challenge-response cycles, citing the user’s own Mirror in each interaction.
• The system enforces anti-sycophancy via code and prompt engineering to avoid praise and ensure pushback based on self-cited content.
• The product is built using technologies including Next.js, TypeScript, PostgreSQL, Supabase, Clerk, OpenAI (GPT-5.6), and Codex.

Note: No evidence of actual users or customers beyond the author’s own use cases is provided.

• The description claims Whetstone offers an “honest thinking partner” — a contrast to typical AI tools that offer only positive reinforcement.
• It positions itself as a tool for helping people discover blind spots in their thinking by challenging assumptions without flattery.
• The author emphasizes that the system prompts are designed to cite users back to themselves, and that this behavior is enforced programmatically rather than relying solely on prompt tuning.
• The product’s core claim — anti-sycophancy — is described as falsifiable and testable through an eval suite.
• There is no indication of a broader market positioning beyond personal development or idea refinement.

Inference: The positioning appears to be niche, focused on introspective or creative individuals seeking honest feedback from AI tools. It does not appear to target enterprise or B2B use cases.

• Not evidenced.
• The description does not identify specific personas, segments, or customer types beyond the author’s personal experience.
• No mention of whether Whetstone targets creators, entrepreneurs, students, or professionals in any domain.

Absence of evidence: There is no clear identification of target customers or ideal customer profile (ICP).

• Not evidenced.
• The description does not contain any information about pricing models, monetization strategies, or revenue streams.
• No indication whether the tool will be free, subscription-based, or offered as a paid service.

Absence of evidence: No business model or pricing structure is described.

• The product was built using Codex for code generation against a locked build spec and guardrails file.
• Authentication went through multiple iterations (magic-link PKCE → token-hash verification → OTP codes → Clerk).
• A test suite evaluates anti-sycophancy via six planted-flaw ideas, asserting no praise openers and citation of Mirrors in first responses.
• The system enforces behavior through code where prompts alone were insufficient — such as mood-based pivots and interview caps.
• Deployment pipeline includes Vercel logs showing 25+ deployments on day one, with eight consecutive failures before a green deployment.
• The architecture uses Next.js, TypeScript, PostgreSQL, Supabase, Clerk, Tailwind, Zod, and OpenAI.

Inference: The engineering approach is highly disciplined, emphasizing spec-driven development and automated testing. However, this does not imply commercial viability or scalability beyond the author’s own use case.

• Not evidenced.
• There is no mention of actual users, signups, retention metrics, or usage data.
• The only evidence of traction is from internal sessions conducted by the author and a few strangers who registered within six minutes.
• No data on customer acquisition, engagement, or product adoption beyond the developer’s own testing.

Absence of evidence: No measurable traction or user behavior data is provided.

• Not evidenced.
• The description does not reference competitors or similar tools in the market.
• No discussion of how Whetstone compares to existing AI thinking partners, idea development platforms, or introspective tools.

Absence of evidence: No competitive landscape or positioning relative to other tools is described.

• The product is described as a solo effort (team size: 1) with no external validation or user feedback.
• It relies heavily on GPT-5.6, which may not be publicly available or stable — especially if it's a proprietary version.
• The system’s anti-sycophancy enforcement is built into code and evals, but there is no evidence of real-world performance outside the author’s controlled tests.
• The product has not been validated with external users or customers, raising questions about its utility beyond the creator’s needs.
• There is no clear path to monetization or scaling.

Inference: The risk of misalignment between the creator's vision and actual user needs is high due to lack of external validation.

1. What specific problems are you solving for users, and how do you know these problems are real?
2. How do you plan to acquire users beyond your own network or personal testing?
3. Are there any external users or beta testers who have provided feedback on the product?
4. What is your long-term vision for monetization or scaling the product?
5. Can you walk us through how the anti-sycophancy mechanism works in practice, and what happens when it fails?
6. How do you plan to maintain quality control as the system evolves beyond the initial prototype?