Travelers brings Claude access to nearly 25,000 employees, citing faster code merges

Travelers is merging code in half the time while giving nearly 25,000 employees access to Claude assistants, according to a new Anthropic case study. The disclosure pairs a wider employee rollout with a specific software-development gain: faster integration of code into Travelers’ environment. It describes an existing deployment, not a rollout starting with this publication.

The starting point was January 2026, when Travelers gave nearly 10,000 engineers, data scientists, analysts and product owners access to Claude assistants. The subsequent expansion puts the reported access figure at roughly two and a half times that initial level. These are access figures, not counts of employees actively using the assistants.

The 50% merge-time reduction is a customer result published by its AI supplier, not an independent evaluation. It measures one step rather than total software delivery time. The case study does not specify the measurement period or comparison baseline, leaving the size of the gain clear but its starting conditions unexplained.

Travelers’ chief AI officer, Matt Gorman, describes the partnership as combining Anthropic’s technology with the insurer’s own resources:

Our partnership with Anthropic is important to how we execute on AI at scale. We pair their leading assistants and models with Travelers’ proprietary data and expertise to help drive our agenda across the enterprise.

Matt Gorman, chief AI officer, Travelers

From early design to code integration

Travelers is integrating AI across the software development lifecycle: discovery, design, building, testing and deployment. Anthropic describes Claude helping teams prototype faster and supporting analysis across the enterprise, extending the deployment beyond writing code alone.

For software teams, the named uses of Claude Code include code generation and analysis, documentation and testing.

Beyond the merge-time reduction, Anthropic also says Travelers teams are writing more code and recording changes more frequently. It gives no numerical increase for either measure.