TL;DR
Since late 2025, the volume of AI-generated code and pull requests has surged, prompting widespread adoption of AI-assisted review methods. Teams are experimenting with layered, risk-based, and automated review strategies, but challenges like noise and trust remain.
Since the end of 2025, the number of pull requests (PRs) on platforms like GitHub has increased fivefold, with growth accelerating sharply. In response, many tech companies are adopting AI-assisted code review tools to handle the surge, with some integrating multiple vendors and layered review processes. This shift is transforming traditional review practices and raising questions about the future role of human reviewers.
Confirmed data from GitHub shows a fivefold increase in PRs over three years, with a near doubling occurring from late 2025. This rapid growth has overwhelmed traditional review workflows, prompting widespread adoption of AI tools such as CodeRabbit, GitHub Copilot Code Review, and custom multi-agent systems. Many teams now rely on AI to generate initial reviews, often reviewing the review itself, with human oversight reserved for high-risk changes. For example, OpenAI and Anthropic implement risk-based approaches, where only high-risk PRs require human review, while low-risk changes are AI-approved for deployment.
Some organizations, like Weaviate, employ layered review loops, where AI agents perform initial assessments, humans decide scope, and AI implements feedback, creating a semi-automated process. Others, such as WeTravel, have refrained from AI review altogether due to noise and false positives. Uber developed uReview, an AI pipeline that filters and categorizes review comments, reducing noise and highlighting critical issues. Despite these innovations, noise remains a significant challenge, with some companies reporting that AI-generated comments can be overwhelming or unreliable.
While the trend toward automation is clear, there is limited evidence of complete abandonment of human review. Even at AI labs like Anthropic and OpenAI, human oversight persists, especially for high-impact or risky changes. The debate continues about whether fully automated reviews are feasible or desirable, with some experts cautioning against over-reliance on AI without sufficient safeguards.
Understanding the Current Trends and Changes in Code Review Practices
AI-generated code has flooded the pull-request pipeline. Since late 2025, engineering teams worldwide have rebuilt their review workflows around AI assistance — layering automation, risk-based approvals, and human judgment in ways that are reshaping software quality culture.
The Surge That Broke Traditional Review
Pull Requests Exploded
GitHub confirmed data shows a fivefold increase in PRs and commits over three years, with growth accelerating sharply from late 2025 as AI code-generation tools became ubiquitous.
AI Reviewers Went Mainstream
CodeRabbit, GitHub Copilot Code Review, and custom multi-agent systems are now common fixtures. AI’s role expanded from code completion into automated review and comment generation.
Human Bottlenecks Emerged
Traditional manual review workflows were overwhelmed. Teams now often review the AI’s review itself, reserving scarce human attention for high-risk changes only.
Pull Request Volume, Relative Scale
Illustrative index based on confirmed GitHub data · fivefold growth over three years, near-doubling from late 2025
The Layered Review Loop
Pioneered by teams like Weaviate, layered review loops create a semi-automated pipeline: AI agents perform initial assessments, humans decide scope, and AI implements the feedback.
🤖 AI Initial Pass
AI agents generate first-round reviews and flag obvious issues automatically.
⚖️ Risk Triage
Changes are scored: low-risk PRs proceed; high-risk ones route to humans.
👩💻 Human Scope Call
Humans decide scope and critical feedback — guarding against scope creep.
✅ AI Applies & Merges
AI implements feedback and merges approved low-risk changes to deploy.
How Organizations Are Adapting
| Organization | Approach | Human Role | AI Role | Key Concern |
|---|---|---|---|---|
| OpenAI | Risk-based approvals | High-risk PRs only | ✓ Approves low-risk | ~ Oversight limits |
| Anthropic | Risk-based + agents | High-impact changes | ✓ Reviews & may merge | ~ Automation trajectory |
| Weaviate | Layered review loops | Scope decisions | ✓ Initial pass + feedback | ~ Agent balance |
| Uber | uReview pipeline | Critical issue focus | ✓ Filters & categorizes | ~ Comment noise |
| WeTravel | No AI review | ✓ Full manual review | ✗ Deliberately excluded | ✗ Noise & false positives |
| Duckbill Group | Near-eliminated review | Minimal oversight | ✓ Risk-based approvals | ~ Overwhelming PR volume |
What Practitioners Are Saying
It’s very hard for agents to get the balance right. Leaving it entirely to AI can cause scope creep or critical issues, so we keep humans in the loop for scope decisions and critical feedback.
Our goal is for another Claude instance to eventually merge low-risk changes, reducing human involvement further.
We recently almost eliminated traditional code review because of the overwhelming number of PRs, relying instead on AI and risk-based approvals.
The Open Debates
Are human code reviews becoming obsolete?
Not yet. Most organizations still involve humans in critical review stages, especially for high-risk or complex changes. Fully automated reviews remain experimental.
How effective are AI tools at reducing noise?
Improved, but imperfect. AI still generates considerable noise, requiring comment filtering, confidence scoring, and human oversight.
What are the risks of AI-only review?
Missed critical issues, scope creep, and false positives — potentially leading to security vulnerabilities and quality problems if unmanaged.
Will AI fully replace human reviewers?
Most experts say human judgment stays essential for high-stakes decisions, at least in the near future, even as automation expands.
What Comes Next
Refined Noise Reduction
Vendors will sharpen comment filtering and confidence scoring to improve review quality and signal-to-noise ratios.
Multi-Agent Maturity
More sophisticated multi-agent systems will handle complex review tasks with minimal human input.
Regulation & Best Practice
Organizational and regulatory standards for AI safety in code review will evolve, shaping industry best practices.
Implications of AI-Driven Code Review on Software Development
The shift toward AI-assisted code review reflects a broader transformation in software engineering, driven by the need to manage exponentially growing codebases and pull requests. This evolution impacts developer workflows, team collaboration, and organizational risk management. While AI can significantly reduce review time and improve consistency, challenges such as noise, trust, and oversight remain. The adoption of risk-based and multi-agent review strategies indicates a move toward more sophisticated, layered review processes that aim to balance automation with human judgment. These developments could influence industry standards, training, and the future role of human reviewers in software teams.
As an affiliate, we earn on qualifying purchases.
Background: Rising Pull Requests and AI Adoption in Code Review
Over the past three years, the volume of code changes submitted via pull requests has increased steadily, with a sharp surge after late 2025. This growth correlates with the widespread adoption of AI code generation tools, which now produce a significant portion of new code. Companies like GitHub have documented this trend, noting a fivefold increase in PRs and commits, with growth accelerating from late 2025. This has prompted a reassessment of traditional code review practices, leading to the integration of AI tools designed to handle the increased workload.
Initially, AI was used primarily for code completion and bug detection, but its role has expanded into automated review and comment generation. Vendors such as CodeRabbit, GitHub Copilot, and custom multi-agent systems have become common. Some organizations have experimented with entirely AI-driven reviews, but most still retain human oversight, especially for critical or complex changes. The debate about replacing human reviewers entirely is ongoing, with many experts emphasizing the importance of human judgment in risk assessment and quality assurance.
“It’s very hard for agents to get the balance right. Leaving it entirely to AI can cause scope creep or critical issues, so we keep humans in the loop for scope decisions and critical feedback.”
— Etienne Dilocker, CTO at Weaviate
As an affiliate, we earn on qualifying purchases.
Uncertainties Surrounding Fully Automated Code Reviews
It is not yet clear whether fully automated, AI-only code review processes will become widespread or remain a niche approach. While some startups and AI labs experiment with removing human oversight entirely, most organizations continue to involve human judgment for high-risk changes. Noise and false positives from current AI tools pose significant challenges, and the long-term reliability and safety of fully automated reviews are still under evaluation. The industry consensus suggests cautious adoption, but the future remains uncertain as AI technology evolves and new standards emerge.
As an affiliate, we earn on qualifying purchases.
Future Developments in AI-Enhanced Code Review Strategies
Expect continued experimentation with layered, risk-based review models, combining AI and human oversight. Vendors are likely to refine noise reduction techniques, such as comment filtering and confidence scoring, to improve review quality. Companies may also develop more sophisticated multi-agent systems that can handle complex review tasks with minimal human input. Regulatory and organizational standards for AI safety and reliability in code review are also anticipated to evolve, shaping best practices. The industry will closely monitor these developments to balance efficiency gains with quality and safety concerns.
As an affiliate, we earn on qualifying purchases.
Key Questions
Are human code reviews becoming obsolete?
Currently, most organizations still involve humans in critical review stages, especially for high-risk or complex changes. Fully automated reviews are experimental and not yet widespread.
How effective are AI tools at reducing review noise?
AI tools have improved significantly but still generate considerable noise, requiring additional filtering and human oversight to ensure quality and safety.
What are the risks of relying solely on AI for code review?
Risks include missed critical issues, scope creep, and false positives, which could lead to security vulnerabilities or quality problems if not properly managed.
Will AI eventually replace human reviewers entirely?
While some industry figures aim for more automation, most experts believe human judgment will remain essential for high-stakes decisions, at least in the near future.
How are organizations managing the increase in pull requests?
Many adopt AI-assisted review, risk-based approvals, and layered review processes to handle the volume efficiently while maintaining quality.
Source: rss
