Warp has added a version of its artificial intelligence (AI) agent for writing code that integrates directly within a command line interface (CLI).
Company CEO Zach Lloyd said rather than working with AI agents within the context of an integrated development environment, Warp Code embeds AI agents with a CLI that is likely to prove more appealing to some developers and many DevOps engineers that prefer a traditional terminal-based coding experience.
Warp Code is based on the same agentic AI framework as Warp, an AI coding tool that is designed to generate code from natural language prompts in a way that provides more control and visibility into how code was developed, said Lloyd. Features such as code review and file editing enables a tighter feedback loop between developer and agent to create better code, he added.
It’s not clear to what degree application developers and software engineers are relying on AI coding tools, but while they improve productivity the code generated is often verbose and inefficient. Additionally, because application developers don’t have a deep understanding of how an AI tool constructed that code, it can be challenging to debug.
More challenging still, most of the large language models (LLMs) that are relied on to generate code were trained using examples of varying quality collected from across the web. As a result, the number of vulnerabilities being generated by AI coding tools is significant.
The only way to address those issues is to create a tighter inner loop workflow that enables developers and the AI agent to more iteratively develop applications in a way that makes it easier for human developers to provide constructive feedback to the AI agent, said Lloyd.
Each organization will need to determine what level of confidence to have in AI coding tools. In some cases, the quality of the code being generated might only be good enough to provide a place to refine it long before it ever makes it to a production environment. However, there are also going to be plenty of instances where the code created by an AI tool might be superior to what a junior or so-called citizen developer might have otherwise generated on their own. At this juncture, it’s not so much a question of whether application developers are going to use AI coding tools so much as it is to what degree.
In the meantime, DevOps teams would be well advised to continuously experiment with and test multiple AI coding tools. The pace of AI innovation continues to accelerate so a tool that only a few months ago might have only been able to reliably execute a few tasks might now be able to manage more complex workflows. The important thing is to make sure to validate the output being generated by these tools because, inevitably, mistakes will be made. The challenge is that far too many application developers are already trusting these tools too much, which ultimately leads to a lot more bad code that has to be, hopefully, weeded out a build before any of it is allowed to make it into a production environment.
