Andrew Finch
Conference affiliation: Staff Software Engineer · Sonar · 2026
Andrew Finch is a staff software engineer at Sonar working on Gitar, its AI code review product. Before working on automated review, he built driver-facing software at Uber and medical records and care delivery systems at Forward, where he led development of the CarePod operating system. His code review examples explore defects that require following an interaction across different code paths, such as a correctly rejected command that leaves the user without a response.
On this page
Driver applications and clinical systems
Andrew worked at Uber from 2014 to 2018, contributing to driver growth and onboarding as well as the mobile app’s on-trip and navigation experience. From February 2017 to July 2018, he was a senior software engineer on the Driver Trip Experience team. His work included navigation and map interactions in the iOS driver app, charging for drivers’ waiting time, and contributions to the team’s rewrite using Uber’s RIBs architecture.
At Forward, from 2018 to 2024, Andrew worked on medical records and care delivery software. He built services that collected and processed members’ health data and helped build tools clinicians used to review that information and order laboratory tests or medication. He also defined the charter for the App Development Platform team and its software development kit, supporting the engineers building those applications. His professional history describes both the clinical applications and the development infrastructure behind them.
Andrew also worked on hardware-software integration and led development of the operating system for CarePod, Forward’s self-contained, AI-powered medical kiosk. This was a distinct responsibility within his broader work on Forward’s medical records and care delivery systems, as described in his updated biography.
Code review across different failure paths
Sonar acquired Gitar in May 2026. Andrew’s work on Gitar includes using the reviewer on his own changes and expressing repository requirements in plain language.
In one pull request, Andrew wrote an access check that correctly blocked free-tier users from invoking a premium command. Through one webhook path, however, the rejected command produced no reply. Gitar’s reviewer caught the missing response. Finding the defect required following what happened after permission was denied: the access restriction worked, but the person who issued the command received no explanation.
For Gitar’s web app, Andrew wrote a Markdown rule for client-side error logging. Client components should call logError whenever they handle a failure. A lint rule looking for catch blocks could cover exceptions, but would miss a server action returning an error as its result or a failed fetch handled in a callback. His plain-language repository instruction expressed one logging requirement that the reviewer could apply across all three patterns.
Andrew’s safe software factory proposal extends this concern beyond individual pull requests. As coding agents increase the volume of changes, he argues for layers of verification across the development process. His own review examples make a specific part of that problem concrete: checking whether a change satisfies a requirement across the paths through which users and failures reach it.