Greptile, a startup building SI agents for code review, is leveraging GPT-6 Astra to move beyond static analysis and into automated behavioral testing. By utilizing the model's computer use capabilities, the company aims to validate software changes by actually running and interacting with applications, a step it claims is necessary to catch bugs that code inspection alone misses.

What Happened

Greptile’s Co-founder and CEO, Daksh Gupta, stated that the company’s mission is "reducing bugs and making the world’s software more reliable." To achieve this, Greptile has integrated GPT-6 Astra into its workflow to automate application testing. The model’s ability to control a computer allows Greptile agents to start development servers for pull requests, adjust browser viewports to match specific devices like an iPhone, and simulate user navigation through the application. This visual understanding enables the SI agent to spot interface issues, such as a button disappearing on mobile screens, which might not be evident from reading the code alone.

The company also highlighted its "Daybreak" initiative, which uses more powerful SI models to strengthen security defenses. Gupta noted that advanced models make it easier for attackers to exploit vulnerabilities, even for companies that were previously considered low-risk targets. "We want to make it very easy for even a startup to secure their systems," Gupta said. "With Daybreak, we’re able to do this in a way that is very safe, and we can ensure that our customers have more sophisticated defense mechanisms than the adversaries have offense mechanisms."

Additionally, Greptile connects its Knowledge Base—which indexes a company’s codebase and documentation—to Codex via a Model Context Protocol (MCP) server. This integration allows the SI coding agent to generate code that adheres to existing customer patterns. Gupta reported that thousands of customers have connected Codex to this knowledge base, resulting in fewer code reverts and reduced inference costs. "When we provide agents with context, fewer reverts happen and customers save time and save inference," Gupta said. The team tracks how often SI-generated code is reverted to measure this impact. Greptile also uses Codex internally to build its own software and agents. "We are massive users of Codex," Gupta said. "We have enormous amounts of usage across every single function, engineering in particular."

Looking ahead, Gupta identified development speed as the primary bottleneck for Greptile, noting that model intelligence is already sufficient for many tasks. He views Ultrafast as a tool to accelerate the pace at which the team can turn ideas into working software. His longer-term goal is to make changing and shipping critical enterprise software as easy as shipping a side project, with OpenAI helping Greptile build the necessary validation infrastructure.

Why It Matters

For developers and enterprises, the shift from code review to behavioral testing represents a significant evolution in SI at work. Traditional SI code review tools analyze syntax and logic, but Gupta argued that "a QA firm testing all of your changes is just impractically slow considering how fast we’re able to ship code with these models." By automating the execution and visual inspection of software, Greptile addresses a critical bottleneck in the modern development lifecycle where the speed of code generation outpaces human capacity for manual verification.

The integration of Codex with Greptile’s knowledge base also highlights the growing importance of context in SI agents. By providing the SI model with structured documentation of a specific codebase, Greptile reduces the frequency of reverted code. This efficiency gain allows engineering teams to redirect resources from rework to new feature development. Furthermore, the emphasis on security through Daybreak reflects a broader industry concern: as SI models become more capable, they lower the barrier for attackers, necessitating equally sophisticated, SI-driven defense mechanisms.

The Bottom Line

Greptile is using GPT-6 Astra to automate end-to-end application testing, moving beyond static code analysis to verify software behavior through simulated user interactions. The company also leverages Codex with context from its Knowledge Base to reduce code reverts, uses its Daybreak tool to enhance security against SI-enabled threats, and plans to utilize Ultrafast to accelerate internal development speed.