Overview
Salesforce adopted Cursor as its primary AI coding tool across 20,000+ engineers, achieving 90% adoption and double-digit improvements in cycle time, code quality, and throughput. The case study details how junior and senior engineers adopted the tool through different paths, and how AI-assisted development is reshaping Salesforce's entire software development lifecycle.
What You'll Learn
1
How a 20,000+ developer organization successfully adopted AI coding tools at enterprise scale
2
Why junior and senior engineers adopt AI coding tools through fundamentally different paths
3
How to measure AI coding tool impact using cycle time, quality, and throughput metrics
4
How AI coding tools can reduce legacy code coverage time by 85%
Prerequisites & Requirements
- Basic understanding of software development lifecycle (SDLC) concepts
- Familiarity with engineering productivity metrics like cycle time and throughput(optional)
Key Questions Answered
How did Salesforce achieve 90% adoption of Cursor across 20,000 developers?
Adoption followed a grassroots pattern where small groups tried Cursor, saw the impact, and the rest followed. Junior engineers adopted first because it helped them understand existing code without in-person mentorship. Senior engineers started with boring, tedious tasks, built trust, then expanded to higher-value use cases. Within a few months, nearly every engineer at Salesforce was using Cursor.
What metrics does Salesforce use to measure AI coding tool effectiveness?
Salesforce evaluates engineering work using three key metrics: cycle time (how fast changes move through the pipeline), quality measured by bug count, and throughput (volume of work completed). They already had dashboards tracking these across thousands of engineers, making Cursor's impact visible almost immediately with double-digit increases in all three areas.
How do junior engineers benefit differently from AI coding tools compared to senior engineers?
Junior engineers, many of whom started careers during the pandemic without in-person mentorship, used Cursor primarily to understand existing code and contribute more effectively. Senior engineers took a different path, first proving out Cursor on tedious, repetitive tasks before building trust and expanding to higher-value use cases as they experienced the tool's reliability.
How much did Cursor reduce legacy code coverage time at Salesforce?
Cursor helped Salesforce reduce legacy code coverage time by 85%. This was part of broader improvements across all three of Salesforce's key engineering metrics—cycle time, quality, and throughput—all of which saw double-digit percentage increases after Cursor adoption.
What challenges remain with AI-assisted code development at enterprise scale?
As more code is written with Cursor's help, Salesforce is still refining how AI-generated code should be reviewed and how to maintain the same level of trust in each change. The company acknowledges they are at the starting point of the AI coding journey and that processes around code review need to evolve alongside the tooling.
What was Salesforce using before Cursor for AI-assisted coding?
Before Cursor, Salesforce had invested in internal AI tools including an open-sourced coding model called Code Genie that thousands of developers relied on. However, Salesforce wanted engineers to have a range of options, so it made Cursor available alongside existing tools, and adoption organically grew to 90% of the engineering organization.
Key Statistics & Figures
Active Cursor users at Salesforce
20,000+
Engineers using Cursor as part of their daily workflow
Cursor adoption rate
90%
Percentage of Salesforce engineers using Cursor
PR velocity increase
>30%
Increase in pull request velocity after Cursor adoption
Legacy code coverage time reduction
85%
Reduction in time needed for legacy code coverage
Engineering metrics improvement
Double-digit increases
Improvements across cycle time, quality (bug count
Company history
25+ years
Duration Salesforce has been building software
Technologies & Tools
AI Coding Tool
Cursor
Primary AI-assisted coding tool adopted across Salesforce's engineering organization
AI Coding Model
Code Genie
Salesforce's open-sourced internal AI coding model used before Cursor
AI Product
Agentforce
Salesforce product shipped with the help of AI-assisted development tools
Key Actionable Insights
1Allow AI coding tool adoption to follow a grassroots, organic pattern rather than mandating top-down rollout. At Salesforce, small groups tried Cursor first, demonstrated value, and the rest followed naturally—reaching 90% adoption within months without forcing adoption.This approach works especially well in large organizations where developers are skeptical of mandated tool changes and prefer peer-validated recommendations.
2Recognize that junior and senior engineers have fundamentally different entry points for AI coding tools and design your rollout accordingly. Junior engineers benefit most from code comprehension and onboarding assistance, while senior engineers should be encouraged to start with tedious, low-risk tasks to build trust before expanding to higher-value work.This dual adoption path was key to Salesforce's success across 20,000 developers with varying experience levels and comfort with AI tools.
3Establish engineering productivity dashboards tracking cycle time, quality (bug count), and throughput BEFORE adopting AI coding tools, so you can objectively measure impact. Salesforce's pre-existing metrics infrastructure made Cursor's value visible almost immediately.Without baseline measurements, organizations struggle to justify continued investment in AI tooling and cannot distinguish genuine productivity gains from perception bias.
4Use AI coding tools to accelerate legacy code coverage and test generation as an early high-impact use case. Salesforce achieved an 85% reduction in legacy code coverage time, which improved overall product quality and reliability.Legacy code coverage is often deprioritized due to its tedious nature, making it an ideal candidate for AI-assisted development that delivers measurable quality improvements.
5Proactively develop new code review processes and trust frameworks as AI-assisted code generation scales across your organization. Salesforce identified that reviewing AI-generated code and maintaining trust in changes is an evolving challenge that requires deliberate process design.As AI generates more code, traditional review practices may not catch the same types of issues, requiring adapted review checklists and potentially new tooling for AI-specific code review.
Common Pitfalls
1
Failing to adapt code review processes as AI-generated code scales across the organization. As more code is written with AI assistance, traditional review practices may not adequately address the unique characteristics of AI-generated changes, potentially undermining trust in the codebase.
Salesforce is actively working on refining how AI-generated code should be reviewed, acknowledging this as an ongoing challenge even after successful adoption.
2
Expecting senior engineers to adopt AI coding tools the same way junior engineers do. Senior engineers need to first prove out the tool on low-risk, tedious tasks before they build enough trust to use it for higher-value work, whereas junior engineers immediately benefit from code comprehension capabilities.
Understanding these different adoption paths prevents frustration and enables organizations to support both groups appropriately during rollout.
3
Mandating top-down adoption of AI coding tools instead of allowing organic, peer-driven adoption. Forcing tool adoption can create resistance, while organic grassroots adoption—where small groups demonstrate value and others follow—leads to more sustainable and enthusiastic usage.
Salesforce's 90% adoption rate was achieved through organic adoption patterns rather than mandated tool usage.
Related Concepts
Ai-assisted Software Development
Developer Productivity Metrics
Enterprise Tool Adoption Strategies
Code Review For Ai-generated Code
Legacy Code Modernization
Sdlc Transformation
Engineering Velocity Measurement
Unit Test Generation With AI