Salesforce ships higher-quality code across 20,000 developers with Cursor
January 21, 2026 at 12:00 PM
Summary
TL;DR
Salesforce rolled out Cursor alongside internal AI tools and reached 90%+ daily use across 20,000 engineers. It reports double-digit improvements in cycle time, bug count, and throughput, plus an 85% reduction in legacy code coverage time.
What actually happened
Salesforce made Cursor available in addition to its internal AI tools
Junior engineers adopted first, using it to understand existing code
Senior engineers started with tedious tasks, then expanded to higher-value work
Adoption spread team-by-team: small group tries it, others follow
AI use expanded beyond coding into more of the SDLC
Key numbers
Salesforce has more than 25 years of accumulated codebase history
20,000 engineers ship new products on top of the codebase
More than 90% of engineers use Cursor daily
Legacy code coverage time reduced by 85%
Double-digit increases reported across key engineering metrics
Why this was hard
A decades-spanning codebase with many system types and legacy constraints
Remote-era junior engineers lacked “learn by pairing” access to senior context
Needed measurable evaluation across thousands of engineers and many teams
As AI-written code grows, maintaining trust in reviews becomes harder
How they solved it
Offered Cursor as an option alongside internal AI, including Code Genie
Let juniors use it to navigate and understand existing code quickly
Let seniors validate value on low-risk, tedious work before broader use
Used existing dashboards to track cycle time, quality (bug count), throughput
Encouraged organic diffusion: early adopters demonstrate impact to the team
Increased unit test generation to raise shipped reliability
What changed
Double-digit improvements in cycle time, bug count, and throughput
Legacy code coverage time dropped by 85%
Engineers generate more unit tests than before
Reported improvement in product quality
Stealable ideas
Start AI adoption with tedious tasks to build trust before expanding scope
Instrument cycle time, bug count, and throughput so tool impact is visible
Support juniors with AI for codebase comprehension when pairing is limited
Treat AI as SDLC-wide tooling, not just a code-writing assistant