CodeRabbit today made available in beta a planning tool that enables DevOps teams to determine which artificial intelligence (AI) prompts to create and review before embedding them within their workflows.

David Loker, vice president of AI for CodeRabbit, said CodeRabbit Issue Planner will enable DevOps teams to collaboratively apply AI to workflows versus relying on a set of individual prompts that have not been properly vetted.

Designed to integrate with Linear, Jira, GitHub Issues and GitLab platforms used by many DevOps teams, the overall goal is to reduce the amount of guesswork that AI agents make in the absence of detailed instructions. That guesswork often results in not only repeated prompting that increases costs, it also results in suboptimal output that DevOps teams need to constantly review, noted Loker.

In fact, a recent CodeRabbit analysis of 470 pull requests (PRs) found AI-authored changes produced 10.83 issues per PR, compared to 6.45 for human-only PRs, with most of those issues involving AI tools being more critical.

What’s needed is an ability for DevOps teams to create a group consensus through which AI prompts are actually validated, said Loker. Today, too many AI agents are being deployed in silos that don’t effectively scale, he added.

As a provider of a code review platform that uses AI to assess the quality of the code being generated by AI coding tools, CodeRabbit has a lot of visibility into how poorly crafted AI prompts can adversely impact DevOps workflows. The challenge and the opportunity is to make it simpler for human software engineers to collaboratively define and review AI prompts rather than having multiple members of a team that have varying levels of AI expertise individually recrafting the same prompts, noted Loker.

While the level of adoption of AI coding tools varies widely from one organization to another, there is little doubt adoption is already widespread. A recent Futurum Group survey, for example, finds a full 60% of organizations are now actively using AI to build and deploy software, with top investments being AI Copilot/AI code tools (38%), AI agent development (37%), AI-assisted testing (37%), DevOps (37%), automated deployment (34%) and software security testing (31%).

It’s not clear how reliable AI agents are just yet, but as the reasoning capabilities embedded in large language models (LLMs) continue to improve so too should the quality of the output. Conversely, however, those same reasoning capabilities have the potential to wreak havoc if the AI agents determine that the best course of action might be to delete a critical database running in a production environment.

Ultimately, DevOps teams will be deploying AI agents that validate the output of other AI agents to ensure tasks are being safely automated. In the meantime, however, the one thing that is clear there is still a need for humans to not just be in the loop, but actually supervise the tasks being assigned to what will soon be a small army of AI agents that will not only have the context needed to create code but also fix it whenever an issue is discovered.