Squire.ai
Summer 2021InactiveNever code alone.
Pressure-test this opportunity
Explore the risks and possibilities with a prompt for ChatGPT, Claude, or your agent.
Never code alone.
Explore the risks and possibilities with a prompt for ChatGPT, Claude, or your agent.
Squire.ai tried to keep developers’ context close to their code. It began as Codex, a tool for explanations inside the editor, passed through EchoLayer’s code-ownership work, and became an AI assistant for pull requests. Its most useful lesson is how a new technology can weaken one product and enable the next. [1] [5]
The outcome needs care. YC currently labels Squire.ai inactive. Its public website and pricing pages remain available, but that does not establish an operating service. Co-founder Brandon Waselnuk’s departure announcement says Saumil Patel and Karl Clement would continue growing the business. The available evidence does not establish a shutdown date, acquisition, revenue, or the reason for the inactive classification. [1] [2] [3] [4]
For a rebuild, another generic review bot faces strong alternatives. A more focused experiment is Verdikt: a queue that helps engineering teams resolve missing or stale review evidence before preparing a change record. That is a proposed workflow, not an established market opening.
Brandon Waselnuk, Karl Clement, and Saumil Patel joined YC’s Summer 2021 batch. The original launch described a team that had built more than 50 products together over eight years. Codex addressed knowledge that disappears when engineers change teams or leave: why a line exists, who understands it, and what an incoming developer should know. [1]
The product placed contextual annotations beside code in VS Code. The intention was to preserve explanations through changes instead of sending developers to a separate document. This was a collaboration tool before it became an AI reviewer. The founders’ prior work together helped explain their choice of developer tools; it does not prove product-market fit. [1]
Codex captured editor context. EchoLayer addressed code ownership. Squire automated small pull-request tasks while leaving responsibility for the complete change with developers. [1] [5]
In his 2024 interview, Patel described agents with different review responsibilities. These included security, code quality, error handling, performance, and scalability. The approach divided a broad review into specialized checks. A description generator could explain a change; a reviewer could identify an issue; another task could help implement feedback. This was the company’s explanation of its architecture, not an independent accuracy benchmark. [5]
The published product site presents automatic reviews, pull-request summaries, team rules, and conversation with a review assistant. Its screenshots show comments inside a pull request and a generated description. Those artifacts make the intended workflow concrete. They do not verify that every advertised capability remains usable today. [2]
Image 1 / 2
Squire says it does not store customer code and has completed SOC 2 Type I and Type II examinations. These are the company’s published statements; no examination report was reviewed here. A buyer still needs current evidence about permissions, retention, model processing, and contractual terms before connecting a private repository. [14]
Squire’s workflow sits where code changes become shared decisions. That is a valuable location: reviewers already inspect a diff, tests, ownership, and discussion. It also creates a distribution problem. A new tool must justify installation, repository access, additional notifications, and a separate bill.
GitHub’s April 2025 launch reported more than one million developers using the review preview in just over a month. That is GitHub’s stated adoption figure, not evidence that Squire lost those customers. Copilot also supports repository context and configuration, so team-specific instructions alone are a weak distinction. [8] [13]
Standalone competition remains substantial. CodeRabbit’s newsroom announces a $143 million Series C at a $1.5 billion valuation in August 2026 and a broader change-management product. Funding does not establish profitability, but it contradicts the claim that independent review products cannot survive alongside GitHub. [9]
Compliance is also an occupied market. Vanta reads GitHub repository settings, pull-request history, and review activity; Drata connects GitHub to code-review and change-management evidence. Verdikt would have to improve an identifiable exception workflow alongside those products. Merely exporting approvals is not a defensible distinction. [10] [11]
Squire’s published pricing lists Teams at $20 per GitHub user per month and Enterprise at $50 per user per month, with a sales call for more than 100 users. These are observed website prices, not confirmed current transactions or historical revenue. [3]
Per-user pricing aligns with an engineering team’s adoption, but the value must exceed both review noise and competing tools’ costs. The public record reviewed here does not disclose paid seats, retention, gross margin, acquisition cost, or cash runway. A directory’s team-size field cannot supply those missing economics.
For Verdikt, the proposed buyer is an engineering operations or security lead handling repeated evidence requests. A repository-based pilot would test whether resolving exceptions saves enough time to earn a separate budget. Any proposed price remains a hypothesis until a buyer pays and renews.
The early beta and waitlist demonstrate interest, not paid conversion. By August 2023, Patel described customers concentrated in the Bay Area as one reason for the planned move. That is a founder account of customer proximity, without a disclosed customer count or revenue. [7]
Squire’s homepage publishes time-saving and merge-speed claims. Without a visible study design, sample, or comparison period, they should be treated as marketing claims. The available record cannot establish a sustained improvement or company-wide commercial scale. [2]
The stronger evidence is product evolution: an original YC launch, a named intermediate product, a founder’s detailed interview, and published review screenshots. Those establish what the team attempted. They leave adoption, economic durability, and the final outcome uncertain.
The documented pressure was technological substitution. Patel said language models reduced the need for the original documentation approach. Codex’s advantage depended on making context easier to capture and retrieve. Once a model could interpret code directly, some of that convenience became easier to obtain elsewhere. The response was to move from preserving explanations toward automating specific tasks. [5]
The commercial challenge changed with the pivot. A context tool asks a team to preserve knowledge. A review assistant asks it to trust recommendations on each change. That makes precision, reviewer effort, and repository access central to the buying decision. This is an interpretation of the workflow, not a reported reason for Squire’s inactive status.
The end of the company is not established. A founder departure, a later competing release, and an inactive directory label do not identify a causal sequence. The record does not support claims that a San Francisco move exhausted funding, that GitHub caused a shutdown, or that all founders abandoned the product. No reliable financial or closure disclosure settles those questions.
The forward-looking risk is clearer. A product built around one model capability can lose its distinction as both platforms and specialists expand. Verdikt should therefore test a concrete operational result: whether a lead can resolve an evidence exception faster and produce a record another person can understand. It should not promise legal compliance or infer AI authorship from coding style.