Proofpoint is a global cybersecurity company providing solutions that help organisations protect people, data and digital communications. Persistent supports Proofpoint across its engineering ecosystem, helping teams improve software development, testing, maintenance and sustenance workflows through AI-assisted engineering capabilities.
Pull request reviews had become a recurring development bottleneck. Developers often had to wait for senior engineers to become available before receiving feedback and progressing code towards production. It also needed to help developers respond quickly to review feedback.
A thorough review required substantial context: the associated Jira ticket and business requirement, pull request description, commits and changed files, test coverage and documentation, coding standards, security and scalability considerations, related feature requirements and design intent and potential impact across connected functions and repositories.
A typical review could take approximately 30 minutes. Quality and depth could vary between reviewers, while repetitive checks consumed senior-engineer time.
Proofpoint needed to complete thorough reviews within minutes, reduce dependence on reviewer availability, bring business, design and QA context into reviews, identify risks, missing tests and standards violations and create a reusable model that could scale across repositories.
Building a Context-Aware PR Review Agent
Persistent built a custom PR Review Agent using Claude Code and Claude Skills. The agent handled repetitive and context-intensive review activity while human reviewers retained final approval.
When a pull request is created, the agent retrieves the PR and relevant project context, reviews the Jira requirement, PR description and code changes and enriches its analysis with Figma, feature and QA context.
It checks for risks, missing tests, documentation gaps and coding issues, then applies a decision framework to distinguish between issues requiring developer action and changes ready for human approval. Detailed feedback is posted to the developer and Microsoft Teams notifications accelerate response.
When developers update the code, the agent can review the pull request again, creating a faster iterative feedback loop before final human approval.
Microsoft’s 2026 engineering guidance shows how quickly AI-assisted review is moving into mainstream software delivery. Microsoft reported that more than 90% of its developers use GitHub Copilot, while AI-assisted code review now covers about 90% of Microsoft pull requests and improves completion time by more than 10%. The broader direction is toward agentic development workflows that span planning, coding, review and delivery rather than treating AI as an isolated coding aid.
For Proofpoint, AI similarly became an initial review layer while engineering judgement remained with people.
Designing the Workflow for Production Adoption
The PR Review Agent was built for live delivery workflows and integrated with tools Proofpoint teams already used, including GitHub, Jira and Microsoft Teams.
Its reusable architecture was adopted immediately within the web-ui repository and is being extended to config-service, risk-service and other repositories.
The agent also formed part of a broader Claude Skills framework. Persistent developed additional Claude Skills for Jira triage and estimation, extending AI assistance across the software development lifecycle.
Software Improvement Group’s State of Software 2026 finds that AI-assisted coding is moving into mainstream enterprise software delivery, but its impact depends heavily on engineering discipline. The report argues that AI amplifies the quality of the underlying software environment: where code and architectural quality are measured and managed, AI can accelerate delivery, while weak controls can increase technical debt, cost and security exposure.
Proofpoint moved beyond isolated AI usage by embedding reusable Skills, project context and human controls into established engineering workflows.
Business Impact
From 30-Minute Reviews to Faster, More Consistent Feedback
- Approximately 70% reduction in pull request review effort through AI-assisted analysis and automated review checks
- Approximately 50% improvement in developer productivity by reducing repetitive review activity and waiting time
- Approximately 40% faster approval cycles through automated reviews, iterative feedback and Teams notifications
- First review pass completed in approximately 8–10 minutes, compared with an average manual review time of around 30 minutes
- An additional quality gate that helped identify risks, missing tests and standards violations earlier
- Immediate adoption within the web-ui repository
- Expansion underway across config-service, risk-service and other repositories
- Three enterprise Claude Skills productionised, supporting automation across more than 150 SDLC activities
The gains were achieved without transferring final approval authority to AI. Human reviewers remained responsible for determining whether changes moved forward.
Building a Reusable Foundation for AI-Assisted Engineering
Persistent helped Proofpoint transform PR reviews from a manual, reviewer-dependent activity into a Claude-powered workflow capable of delivering thorough feedback within minutes.
By combining Claude Skills with Jira, Figma and QA context, configurable review modes, automated Microsoft Teams notifications and human-led approval, the solution improved review consistency while accelerating developer response and merge readiness.
With immediate adoption in web-ui, expansion across additional repositories and three Claude Skills supporting more than 150 SDLC activities, Proofpoint now has a reusable foundation for scaling AI-assisted engineering productivity across its software development lifecycle.
Accelerate Engineering Reviews. Scale Context-Aware AI Workflows. Talk with Persistent.
