Pythagora AI today made generally available an artificial intelligence (AI) coding tool that is designed to more interactively enable application developers to automate the writing of code and streamline debugging workflows.
Company CEO Zvonimir Sabljic said that the approach ensures that developers are able to maintain the appropriate level of context needed to fix and optimize code the AI tool generates.
That’s crucial because it makes it simpler for developers to debug code they are familiar with versus relying on a tool that generates code in a way that a developer doesn’t really understand how it was constructed, he added.
While there has been widespread adoption of AI coding tools, a lot of that code is not able to be successfully deployed in a production environment. The core issue is that AI tools have been trained using samples of code pulled from across the web rather than using code that has already been deployed in a production environment. As a result, much of the code generated by the tools still needs to be debugged by application developers before it can be deployed.
While the Pythagora AI platform has been similarly trained, the workflow is designed in a way that ensures developers are more engaged as code is created, thereby making it easier for them to debug that code when necessary, said Sabljic. In effect, rather than simply generating code the Pythagora AI platform is designed to provide application developers with a partner that helps build applications alongside them by asking for feedback at every step of the workflow, he added. That capability ensures that all the most important coding decisions are still made by a human, said Sabljic.
At the core of the Pythagora AI platform is GPT Pilot, an open-source tool that engages application developers in conversations to generate code, including providing explanations of how code was created. Application developers can then review that code and manually customize it as they see fit, noted Sabljic. The goal is to ensure that developers eventually determine it is taking them longer to debug unfamiliar code than it would to write it themselves, he added. Once approved, it can be added to a code base with a single click, noted Sabljic.
There is already no shortage of AI coding tools and each application developer will need to decide on the degree they wish to rely on them. The one certain thing is thanks to these tools the overall amount of code being generated has increased. How much of that code is making it into production environments, however, is less apparent. Nor is it clear to what degree all that increased code is leading to more applications being built and deployed faster when less than 20% of the development effort actually involves writing code.
Regardless of how AI coding tools are ultimately employed, there’s no going back to manually writing every line of code. The challenge and the opportunity now is finding that middle ground that enables developers and machines to collaboratively build and deploy applications in a way that involves much less toil than it does today.