Since the end of 2025, the landscape of software development has shifted dramatically, with AI-generated code contributing to a fivefold increase in pull requests on platforms like GitHub. This surge has prompted companies to overhaul traditional code review processes, integrating AI tools at various stages to cope with the deluge of code changes. While some teams rely heavily on AI for reviews, others maintain a mix of automated and manual checks, reflecting a landscape in flux.

The primary driver behind these changes is the rapid growth in code submissions, with GitHub data showing a fivefold increase in pull requests over three years, accelerating sharply after late 2025. This growth is largely attributed to AI agents generating code faster and in larger volumes than human developers, leading to an urgent need for scalable review methods.

Many organizations now incorporate AI review tools such as CodeRabbit, GitHub Copilot Code Review, and Claude Code Review, which automatically analyze pull requests, leave comments, and sometimes even suggest fixes. Some teams employ multiple AI agents in a multi-agent review process, where different models evaluate the same code and propose modifications. These AI reviews are often supplemented by human oversight, especially for high-risk changes, with the goal of balancing speed and safety.

Another common approach involves triaging code changes based on their ‘blast radius’ — low-risk modifications are approved with minimal or no human review, while high-risk updates undergo thorough manual inspection. Companies like Anthropic and OpenAI follow this risk-based review model, with plans to automate even low-risk merges in the future. Additionally, some teams focus on smaller PRs to make reviews more manageable, or set up AI agents to update code automatically, reducing reviewer workload.

Despite these innovations, challenges persist. Noise—irrelevant or low-confidence comments from AI—remains a significant issue, leading some companies, such as WeTravel, to avoid AI reviews altogether. Others, like Uber, have developed sophisticated pipelines like uReview, which grades and filters AI comments to highlight only the most critical feedback. There is also ongoing debate about the extent to which human review can or should be replaced, with many organizations still insisting on manual checks for critical or high-risk code.

Implications of AI-Driven Code Review Strategies

The shift toward AI-assisted code reviews impacts development teams by increasing review capacity and speed, but also raises questions about reliability, trust, and quality assurance. Automating parts of the review process can reduce bottlenecks, but it may also introduce new risks if AI comments are ignored or misunderstood. For organizations handling sensitive or complex codebases, maintaining human oversight remains essential to prevent critical errors and ensure compliance.

Furthermore, the evolving landscape influences organizational workflows, requiring teams to develop new protocols for risk assessment, AI tool integration, and human-AI collaboration. The move toward reduced human involvement in reviews could reshape team structures, roles, and responsibilities over time, emphasizing the need for clear guidelines and trust in AI systems.

Growth of Pull Requests and AI Adoption Since 2025

Over the past three years, the volume of pull requests on GitHub has increased fivefold, with a sharp acceleration after late 2025. This surge correlates with the widespread adoption of AI code generation tools, which enable developers to produce code faster and in greater quantities. The increase in pull requests has created a bottleneck for traditional review processes, prompting organizations to seek scalable solutions.

Leading AI companies, such as Anthropic and OpenAI, have integrated AI review systems into their pipelines, adopting risk-based approaches to prioritize human review for critical changes. Startups like Duckbill Group have even scaled back or eliminated manual reviews, relying heavily on AI to manage their codebase. Meanwhile, industry research indicates that noise and trust issues remain significant hurdles, with some firms opting to avoid AI reviews altogether due to concerns over irrelevant comments and false positives.

Overall, the data and industry practices paint a picture of a rapidly evolving review landscape, driven by technological advances and the need to handle exponentially increasing code submissions efficiently.

“It’s very hard for agents to get the review balance right. I favor a setup where an adversarial agent reviews, a human makes scope decisions, and AI implements feedback, with humans overseeing critical steps.”

— Etienne Dilocker, CTO at Weaviate

Uncertainties Surrounding AI-Only Code Reviews

It remains unclear how widespread or permanent the shift toward AI-only or minimal human review will become. While some startups experiment with eliminating manual reviews entirely, there is limited evidence of broad adoption at scale. Concerns about AI reliability, noise, and the potential for overlooked critical issues continue to hinder full automation.

Additionally, the long-term impact on code quality, team dynamics, and organizational safety protocols is still being studied. The balance between speed, safety, and trust in AI systems remains a key point of debate among industry leaders and researchers.

Future Developments in AI-Enhanced Code Review

Expect continued experimentation with AI review tools, including more sophisticated multi-agent systems and better noise filtering. Companies will likely refine risk-based approaches, automating low-risk merges while maintaining manual oversight for complex or sensitive code.

Further research and industry best practices are anticipated to emerge around AI trustworthiness, explainability, and integration standards. Over the coming months, more organizations will publish case studies and benchmarks, clarifying the effectiveness and limitations of AI in code review workflows.

Ultimately, the evolution of code review practices will depend on technological advances, regulatory considerations, and organizational risk appetite, shaping a new norm for software development in the AI era.

Key Questions

Are all companies replacing human code reviews with AI?

No, most organizations are adopting hybrid approaches, combining AI tools with human oversight, especially for high-risk or complex changes.

What are the main challenges of AI-based code review?

Key issues include noise from irrelevant comments, trust in AI recommendations, and the risk of missing critical errors without human review.

Will AI eventually fully replace human reviewers?

It is uncertain. While some startups experiment with minimal or no human review, many experts believe human oversight remains vital for safety and quality assurance.

How are teams managing the increase in pull requests?

Many are implementing AI review tools, risk-based triage, smaller PRs, and automation pipelines to handle the volume efficiently.

Source: rss