Driving the Agentic Shift in Salesforce Engineering Practices

Salesforce Transforms Software Development with AI-Driven Tools

Salesforce has made significant strides in integrating artificial intelligence (AI) into its software development processes, transitioning from traditional methods to an AI-driven approach that enhances productivity and quality. The company has shifted its focus to using Claude Code as its primary AI agent, leading to remarkable improvements in the software development lifecycle (SDLC). This transformation was marked by a notable increase in work efficiency and output quality, showcasing the potential of AI in modern engineering practices.

The Shift to Claude Code

The pivotal moment for Salesforce’s engineering team came with the decision to adopt Claude Code as their main AI tool across the organization. This strategic move was complemented by the removal of all token limits associated with the tool, effectively eliminating barriers that previously hindered engineers’ productivity. By streamlining access to AI capabilities, Salesforce aimed to empower its developers and enhance their effectiveness.

The impact of this transition is evident in the data. In April 2026, Salesforce reported a 50.8% increase in work items completed per developer compared to April 2025. Additionally, the number of pull requests (PRs) merged per developer surged by 79%. More importantly, when evaluating the actual value of code delivered through a machine learning-based Effective Output score, there was an impressive year-over-year growth of 151.3%. These metrics underscore how the integration of Claude Code has not only accelerated development but also improved overall output quality.

Agentic Transformation in Action

While statistics provide a glimpse into Salesforce’s progress, real-world examples illustrate the profound changes taking place within the organization. One notable instance involved a product team tasked with migrating 33 API endpoints to a new cloud-native architecture—a project traditionally estimated to require around 231 person-days. By leveraging Claude Code and implementing a rule-based framework for automation, the team completed this migration in just 13 days—an astonishing eighteen times faster than expected.

This rapid completion was achieved through innovative strategies such as incorporating feedback from each round of PRs back into their rule set, allowing for continuous improvement in accuracy and output readiness. The team utilized autonomous large language model loops to manage tasks without manual intervention while parallelizing migrations across isolated environments, resulting in multiple PRs being generated simultaneously. This approach not only expedited the process but also showcased a fundamentally different method of software development.

Balancing Output and Quality

A common concern when implementing AI at scale is whether increased speed comes at the expense of quality. However, Salesforce’s Engineering 360 platform provides compelling evidence that this is not the case. Despite a significant rise in PRs, total incidents dropped by 5%, indicating that productivity gains did not compromise quality standards.

This finding challenges the conventional notion that productivity and quality are often at odds. Instead, Salesforce emphasizes trust as a core value, ensuring that engineers leverage their AI capabilities without sacrificing high-quality standards or security measures. By embedding security protocols and quality benchmarks directly into their workflows, Salesforce demonstrates that rapid development can coexist with rigorous quality control.

Rethinking Engineering Workflows

With widespread adoption of AI tools like Claude Code, Salesforce engineers are now reimagining traditional workflows within the SDLC. The focus has shifted towards identifying unnecessary processes and handoffs that can be eliminated entirely or streamlined through automation. This introspection encourages teams to consider how much human intervention is genuinely required versus what can be delegated to agentic systems.

As engineers gain experience with these new workflows, they are evolving from mere users of AI tools to architects of their own agentic systems. Skills related to creating reusable capabilities—termed Claude Code skills—are becoming essential artifacts within engineering teams. These skills encapsulate team-specific contexts and workflow patterns that can be shared and built upon collaboratively.

Navigating Challenges Ahead

Despite these advancements, several challenges remain for Salesforce as it continues its journey toward an automated SDLC. Engineers are still grappling with context management during lengthy agentic sessions; inconsistencies in CLAUDE.md files—the configurations guiding Claude’s interactions—can significantly impact output quality.

Security also poses unique challenges in an environment where agents can execute tasks autonomously rather than merely suggesting actions. As such, Salesforce is investing heavily in securing its agentic SDLC comprehensively.

The evolution of roles within engineering teams is another area requiring careful consideration. As agents take on more execution responsibilities traditionally held by junior engineers, questions arise regarding how these roles will adapt over time and what new opportunities will emerge for designers and product managers within this changing landscape.

What This Means

The transformation underway at Salesforce illustrates a broader trend toward integrating AI deeply into software development practices. By embracing tools like Claude Code and fostering an environment where engineers can build their own workflows, Salesforce is setting a precedent for future engineering organizations. The lessons learned from this journey highlight not only the potential for increased productivity but also emphasize maintaining high-quality standards amidst rapid change—an essential balance for any technology-driven enterprise aiming for sustainable growth.

For more information, read the original report here.