Faros.AI

Here is a fact that should bother more executives than it does: engineering is usually the single most expensive line item in a software company, and it is often the least measured. Sales has a pipeline you can stare at all day. Finance has a burn rate down to the dollar. Marketing has attribution models that would make a physicist blush. Engineering - the department spending all the money - frequently runs on vibes, standup meetings, and a VP's gut feeling about whether things are "on track." Faros AI exists to fix that specific, expensive gap.

The company was founded around 2020 by three people who had just spent years building Salesforce Einstein, one of the first serious enterprise AI platforms. Vitaly Gordon had been a VP of Engineering at Salesforce, and he noticed something odd while running large teams: he could tell you almost nothing about how his own organization worked, quantitatively, even though he was drowning in tools that generated data all day long. GitHub knew who committed what. Jira knew which tickets moved. The CI system knew what deployed and what broke. But nobody had connected the dots, so the person responsible for the whole machine was, in an analytical sense, flying blind. That observation is the entire company.

What Faros built is, at its core, a data infrastructure layer. It ingests from more than 100 sources - version control, issue trackers, CI/CD, incident management, calendars, HR systems - and normalizes all of it into one connected model of how engineering work actually moves. On top of that sits the part customers see: dashboards for DORA metrics, delivery and velocity views, program tracking, and increasingly, measurements of what AI coding assistants are doing to all of the above. The pitch is not "here is another dashboard." The pitch is "here is the honest single source of truth underneath all your dashboards."

The paradox they put a number on

The most interesting thing Faros has done recently is not a feature - it is a finding. As every engineering org rushed to hand developers AI assistants, a natural and slightly nervous question emerged: is this actually working? Faros went into the data and surfaced what it calls the AI Productivity Paradox. Individually, developers using AI produce dramatically more: roughly 21% more tasks completed and 98% more pull requests merged. Individually, this looks like a revolution. And yet, at the organizational level, delivery metrics frequently stay flat. The extra output piles up somewhere - in review queues, in context-switching, in work that moves faster into a bottleneck that did not move at all.

This is exactly the kind of result that is easy to feel and hard to prove, which is why it matters that someone measured it. If you are a leader who just spent a seven-figure budget on Copilot licenses, "our developers feel faster" is not a number you can take to a board. "Individual throughput is up 98% but our cycle time is unchanged, and here is the specific stage where the work is getting stuck" is a number you can actually act on. That is the product, dressed up as research.