Gonfire

Gonfire is an assessment platform designed to evaluate how candidates interact with AI tools when working on real codebases. It addresses the challenge of distinguishing genuine engineering skill from AI-generated outputs by capturing and analyzing every interaction a candidate has with AI during technical assessments. The platform is built on the premise that the process by which a candidate steers AI assistance is more informative than the final code submission, which often appears similar regardless of the candidate's input due to widespread AI usage.

The tool provides a detailed record of the candidate's workflow, including prompts issued to AI, edits made to the code, reverts, test runs, and decision points such as accepting or reviewing AI suggestions. Rather than focusing on labeling individual lines as human- or AI-written, Gonfire emphasizes behavioral signals—tracking which prompts led to which decisions and where candidates engaged critically with AI output. The assessment includes a full interaction trail, allowing reviewers to observe planning, prompting, and recovery steps in context. Candidates are explicitly informed that their AI interactions will be recorded and evaluated as part of the process.

Gonfire supports assessments using any GitHub repository or zip file as the starter source, with tested compatibility for TypeScript, Python, Rust, and Go. The setup process is designed to be minimal for both hiring teams and candidates, requiring only a repository URL and a brief. The platform is intended for use by engineering leaders and hiring teams seeking to understand not just what candidates deliver, but how they approach problem-solving with AI tools. Gonfire does not aim to replace the entire interview process, but to supplement or replace traditional take-home coding assessments by providing insight into the candidate's workflow rather than just their output.

At present, Gonfire is in early access and is being calibrated with a small set of design partners. Pricing information has not yet been published, and access is available by request while the product is refined in collaboration with early users.