Greptile Automates Application Testing with GPT-6 Astra

Greptile is betting on GPT-6 Astra, Codex, and Ultrafast to automate application testing in real-world conditions and accelerate software delivery. Its CEO, Daksh Gupta, claims that providing context through a knowledge base reduces code rollbacks and saves time, while Daybreak aims to enhance customer security. The company also aspires to simplify the delivery of critical software.

Ultrafast Presented as a Lever Against Bottlenecks

For Greptile, Daksh Gupta considers that the main obstacle is the speed at which the team transforms ideas into functional software. He believes that the intelligence of current models is sufficient for many tasks and presents Ultrafast as a means to accelerate this process. He expresses particular enthusiasm for advancing at the pace of good ideas. In the long term, he aims to make the change and delivery of critical software as simple as shipping a side project, with the help of OpenAI to build the necessary validation.

Automating User Testing Experience with GPT-6 Astra

Greptile utilizes GPT-6 Astra to automate large-scale application testing. According to Daksh Gupta, the available computing power allows for simulating the experience of a user attempting to break an application. The validation of a pull request includes checking logic, developer intent, security, and non-regression, with some checks requiring the actual execution of the application. Daksh Gupta gives the example of a button that disappears on an iPhone: Greptile can launch a development server, adjust the browser to an iPhone screen size, and simulate navigation, with the model's visual understanding enabling it to spot the hidden element. He believes that a QA company testing all changes would be too slow compared to what these models allow.

Reducing Rollbacks Through Context Provided to Codex

Greptile's knowledge base indexes clients' source code and documents its functionality in a format usable by both humans and agents. This documentation is accessible to Codex via a Model Context Protocol server, allowing for the generation of code that complies with clients' existing models. According to Daksh Gupta, thousands of clients have connected Codex to this knowledge base. The team tracks the frequency of code rollbacks generated by AI and observes that with context, these rollbacks decrease, saving time and inference resources. Clients can thus allocate these resources to new developments. Greptile also uses Codex for its own tools and agents, with particularly strong usage in the engineering field. The company designs artificial intelligence agents tasked with analyzing and validating code changes while considering the entire source code, a task that Daksh Gupta associates with reducing errors and improving reliability, while emphasizing the need to evaluate application behavior.

Security: Daybreak Claims Superior Defenses Against Attacks

Daksh Gupta states that the rise of models also makes it easier for attackers to exploit applications, exposing more businesses. He wants to make security accessible even to startups. With Daybreak, he claims to be able to protect clients' systems very securely and ensure that their defense mechanisms are more sophisticated than those of their adversaries.