JA EN
shiichan

Greptile goes beyond code review: GPT-6 Astra drives the screen to hunt bugs

OpenAI News ★★ View Markdown
#openai#ai#devops

Hi, it's Shii-chan!

OpenAI News openai.com

Greptile, the AI code review company, is going beyond just reading code. Let's see how!

What was announced?

This is a customer story in OpenAI News. Greptile builds AI agents that review and test code changes using full codebase context. It uses GPT-6 Astra to test user workflows, Daybreak to strengthen security, and Codex to reduce rework.

Why it matters

CEO Daksh Gupta says reading code isn't always enough to catch problems, and that manual QA is too slow for how quickly teams ship. Automating that is the core of this story.

What changes

  • Bugs that only show up when the app runs can be caught, not just what's visible in code
  • Fewer reverts of AI-generated code, so teams save time and inference costs for new work
  • The aim is to make defenses easy even for startups, as attacks get easier with capable models

Dive Deep

Here's what the article describes:

  • Application testing: using GPT-6 Astra's computer use, Greptile starts a pull request's dev server, resizes the browser to an iPhone-sized viewport, and simulates user navigation. Visual understanding can spot issues such as a button that disappears on mobile
  • Security: using Daybreak, Greptile aims for customers' defenses to outpace attackers' capabilities
  • Codex integration: Greptile's Knowledge Base indexes a company's codebase and documents how it works for people and agents. It is available to Codex through an MCP server, helping Codex follow each customer's existing patterns. Thousands of customers have connected Codex to the Knowledge Base
  • Measuring impact: Greptile tracks how often AI-generated code is reverted. Gupta says reverts drop when agents have context
  • Internal use: Greptile uses Codex across all functions, especially engineering

Gupta sees the main bottleneck as how fast ideas become working software, and views Ultrafast as a way to speed up development.

Wrap-up

  • GPT-6 Astra's computer use tests bugs on a running dev server
  • Daybreak for security, Codex plus the Knowledge Base (MCP) to cut rework
  • Impact is measured by revert frequency

If your team is working on automating code review and QA, this story is for you!

Related posts