Why Crossfire exists.
Crossfire started as a working method before it was a product. The same review prompt went to three models, the answers were compared, and where they agreed we moved on. Where they split, we argued it out. Crossfire's own product requirements went through five versions that way, and its technical design through seventeen.
The method worked because the count was informative. One model's confident answer was easy to accept. Three models with two positions forced a real decision, and three with one position let us stop second-guessing. Crossfire turns that method into a tool for people whose documents have to be right.
How it is built
Three codebases: a Python backend, a React web workspace and a Microsoft Word add-in, covered by roughly 3,000 automated tests.
The code is written by AI coding agents working in parallel, each with one job. Gemini scopes and judges the architecture. Codex picks up issues and implements them. Claude builds the interface, audits every change, and closes the books. Linear is the one record of what was asked, what was built, and what was checked. A human sets direction and makes the calls the agents are not allowed to make.
The product reviews its own designs with the same multi-model method it sells. Its requirements went through 5 versions that way, and its technical design through 17.
Who
Crossfire is built by Chad Houghton. By day he is a director of technical product management at Verint, where he owns an 11-product portfolio used across 8 million licensed seats and works alongside an 85-person engineering group.
Before product, he ran DevOps and cloud infrastructure, and shipped software at Intuit and Amazon Games. He built Crossfire because he was already reviewing documents with three models by hand, and the part that took the longest was comparing the answers. He lives in San Diego. Find him on LinkedIn.