Ramp spent $1.5 million on AI model tokens in a single week in May. Then it built a product to get more out of every dollar.
The corporate spend management company has launched Glass, an internal AI workspace that turns company-specific knowledge and workflows into reusable tools. CEO Eric Glyman announced the release on October 13, 2026, and early access signups are open through Ramp’s website.
What Glass actually does
In the AI world, tokens are the small chunks of text that large language models read and write, and providers bill customers based on how many get processed.
Glass is meant to channel that usage into something more organized. A workflow that works once gets saved, packaged and reused by everyone else, rather than reinvented from a blank prompt every morning.
The platform sets itself up automatically through Single Sign-On (SSO), and plugs into the internal tools Ramp already uses. Its interface is a split-pane workspace, so users can see the AI and their work side by side.
The headline feature is a marketplace holding more than 350 reusable skills. Rounding things out are automation and persistent memory, which holds onto context between sessions so users do not have to re-explain the same background every time they open a new chat.
Glass has reportedly cut Ramp’s AI costs by 30% in certain accounts.
A company already deep in the AI pool
Ramp reports that 99.5% of its workforce has adopted AI internally.
The company’s internal software factory, called Inspect, is responsible for more than two-thirds of Ramp’s code development, putting it north of 66% of the codebase.
Ramp has also expanded its services to include tools that help businesses manage AI token spending across multiple providers, including OpenAI, Anthropic and Gemini.
Ramp’s recent funding rounds have valued the company somewhere between $44 billion and $60 billion.
What this means
The 30% cost reduction is reported for certain accounts, not across the board, so outside companies should not assume identical savings. Glass began as an internal workspace, with access currently offered through an early signup, so broad real-world results are still ahead.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.