In a revolutionary technical move aimed at bringing about a fundamental change in the way engineers and developers work around the world, artificial intelligence has entered a new phase of professionalism and precision by auditing and analyzing code simultaneously and complexly to detect errors.
- Introduction to the new code review feature
- Collaborative workflow of the five digital agents
- Smart filtering system and reducing noise for developers
- Focus on cybersecurity and historical vulnerability detection
- Intense competition and pricing in the software development market
- Frequently asked questions
Introduction to the new code review feature
Leading emerging technology company Anthropic announced on Monday the launch of a completely new and innovative software feature called “Code Review,” integrated directly as a qualitative and powerful addition to its popular developer tool “Claude Code.” This modern and exceptionally capable technology relies on deploying and activating multiple AI agents simultaneously and in parallel to perform automated and deep auditing of pull requests, with the primary goal of discovering hidden bugs and complex security vulnerabilities that could threaten projects. The company clearly stated in its official statement that this innovative system was carefully designed and developed to catch “bugs and vulnerabilities that are typically ignored or overlooked by human reviewers” due to work pressure, fatigue, or code complexity and intertwining. This powerful feature is currently available as a research preview for team plans and enterprise customers, with a single typical and comprehensive review taking about 20 minutes to complete, and costing between $15 and $25 USD per review depending on token consumption volume and required processing power, according to financial website StreetInsider.
Collaborative workflow of the five digital agents
The new system’s workflow features deploying and directing 5 completely independent AI agents that analyze and audit changes to code from different and contrasting perspectives in real time. Each specialized digital agent focuses on a specific and vital aspect of deep technical review, as follows: the first agent is dedicated to checking compatibility and compliance with “Claude” model instruction file standards, the second agent’s job is to discover deep structural and logical bugs, the third analyzes the context of past change logs to understand interrelatedness, the fourth evaluates previous comments associated with the pull request to avoid redundant work, and finally the fifth agent handles the accurate verification of the relevance and correctness of explanatory comments written within the code itself by the engineer. To ensure the highest degrees of accuracy and avoid false positives that annoy developers and disrupt daily workflows, the agents collectively evaluate each issue they discover and assign it a unified confidence score on a precise scale ranging from 0 to 100.
Smart filtering system and reducing noise for developers
The tool’s capabilities and intelligence do not stop at merely monitoring and detecting technical errors, but extend to managing workflows in an organized and quiet manner that does not distract attention. The system is precisely programmed not to publish warning comments and feedback to development teams unless the discovered issue exceeds the default minimum confidence threshold strictly set at 80 points—a smart engineering design aimed primarily at reducing unnecessary noise and saving valuable engineering time. The advanced system also features smart filtering capabilities that automatically exclude and skip pull requests that do not require or need actual review to save resources, such as closed pull requests, incomplete drafts, automated requests generated by systems, or those previously reviewed. Furthermore, the feedback and notes included by the tool feature direct and immediate links to GitHub, accompanied by precise review code hashes and specific line number ranges, making it easy for developers to navigate directly and quickly to the error location and start addressing it immediately without searching hassle.
Focus on cybersecurity and historical vulnerability detection
The launch of this advanced feature comes as the culmination and continuation of a continuous series of updates focused on meeting developers’ security and operational needs added to “Claude Code” over the past few months. Last February, Anthropic introduced a separate and highly effective security capability called the Security Tool, dedicated to scanning entire codebases for critical vulnerabilities and suggesting necessary patches to remediate them instantly. This tool, powered by the latest “Claude Opus 4.6” model, helped the company’s internal team discover and identify more than 500 real and critical security vulnerabilities in open-source codebases used in production environments worldwide. The company noted that these complex bugs remained undiscovered and present for many years despite undergoing frequent reviews by professional human experts. Earlier in February, the company also launched other important features including live application previews, cross-platform pull request monitoring with auto-fix capabilities, and a voice command mode to facilitate direct interaction.
Intense competition and pricing in the software development market
This new and advanced release arrives at a time when the technological arena is witnessing intense and fierce competition in providing AI-generated code review services. Independent developers and engineers had already begun building and customizing their own multi-agent review workflows leveraging Claude Code’s sub-infrastructure, with some advanced setups created by the developer community running and directing up to 9 parallel review agents simultaneously to ensure maximum protection. Strong market competitors, such as OpenAI’s Codex model and GitHub’s Copilot programming assistant, also offer automated review features seamlessly integrated into developer workflows through recurring subscriptions. Anthropic’s unique approach of charging a financial fee for each individual review instead of bundling the feature into a fixed monthly subscription reflects the intensive consumption of energy and heavy computing resources required to run multiple AI agents in parallel. This cost structure is likely to play a major role in how engineering teams evaluate economic feasibility and decide when comprehensive automated review is worth this additional financial expenditure.
Frequently asked questions
Question: What is the core technology behind Anthropic’s new code review tool?
Answer: The advanced tool relies on running 5 AI agents in parallel and independently to analyze code from multiple angles to discover vulnerabilities with high precision and reliability.
Question: How much does a single code review cost and how long does it take to complete?
Answer: The review costs between $15 and $25 USD based on the token volume consumed by the code, and the complex process takes about 20 minutes to fully complete.
Question: How does the tool prevent annoying developers with false and repeated alerts and reviews?
Answer: The tool evaluates each error and assigns it a confidence score, and does not publish any alerts or notes to teams unless it is more than 80 points confident that the error exists on the scale.
Question: What is the important security achievement accomplished by the company’s tools in previous updates?
Answer: A previous security tool launched by the company managed to discover more than 500 vulnerabilities in open-source software that remained hidden for years despite intensive human reviews.
