Daksh Gupta works on AI systems for software reliability at Greptile. On his personal website, he describes the work as building agents that autonomously review code and catch bugs before production. He studied computer science at Georgia Tech, according to his AI Engineer World’s Fair 2026 biography, which listed him as Greptile’s co-founder and CEO. That conference biography described Greptile as reviewing code changes for more than 7,000 companies, including Nvidia, Coinbase and Scale, and reported that the company had raised $30 million from Benchmark, Y Combinator, Paul Graham and others.
His work centers on understanding whether a proposed software change will behave correctly within an existing application. Greptile’s website describes a system that builds a graph of files, functions and dependencies, then uses agents to examine pull requests and their effects beyond the changed lines. It also describes runtime validation that runs a pull-request branch in a sandbox to find bugs that appear during execution. In his AI Engineer presentation, Gupta explains how agents inspect changed and related files, install dependencies, start applications, supply mocked inputs and use browser agents to exercise running software.
Gupta has also examined how agent-written code compares with human-written contributions in enterprise codebases. In that presentation, he describes analyzing Greptile’s pull-request data using author fields, co-author footers and branch-name prefixes to identify likely AI-generated changes. He reports broadly similar results for human and agent contributions across revert rates, comparisons involving pull-request size, the severity of flagged issues and review rounds before merge. His analysis also explores differences in failure patterns, including SQL injection and authentication bypass, through the language of review comments. These findings are observational comparisons within Greptile’s data.
As coding agents increase the volume of proposed changes, Gupta argues that validation must establish whether a change breaks behavior users rely on, makes future failures more likely or accomplishes its author’s intent. His approach connects codebase analysis with testing running applications, addressing the practical challenge of reviewing and safely deploying software as code production accelerates. AI Engineer presentation
Daksh Gupta examines Greptile’s enterprise pull-request data, finds similar quality signals for human and agent code, and explains why growing code volume changes what validation must accomplish.
Agent authorship needs multiple signals. Author fields alone identified fewer than 1% of PRs; adding co-author footers and branch prefixes raised the estimate to roughly a quarter, with full autonomy still inferred.
Greptile’s observational comparisons found broadly similar human and agent results across reverts, PR-size analysis, flagged issue severity, and review rounds before merge.
Overall quality can conceal distinct failure patterns. Comment-based comparisons put Claude’s SQL-injection frequency at about 1.5× the human baseline and Devin’s authentication-bypass frequency at about half.
Validation should examine existing user behavior, future failure risk, and author intent. Greptile combines inspection of changed and related code with sandboxed browser interaction to support that work.
▶ Watch ↗