Workday makes Sana the home screen with TypeSafe's Jev decision model

Sana becomes the home screen

Workday on Tuesday made Sana, its AI assistant, the new home screen for HR, finance and IT software. The interface is built on the fly, managed by Jev, TypeSafe's decision model. In a demo, a paycheck question brought up a card with the amount hidden behind a click, while a time-off request produced a date picker, with Sana filling in the rest. Workday bought Sana for $1.1 billion last year. It is available now, and admins can roll it out without making it the default.

Choosing components instead of writing them

Joel Hellermark, Workday's chief AI officer and Sana's founder, said the usual generative UI approach, in which a model generates the full interface, is "quite slow." He said a classification model picking design components can work instantly. Workday gave Jev every atomic component in its design system and hard-coded none of the selection logic, though data preloading mattered. TypeSafe says Jev answers in 70 to 500 milliseconds. Hellermark said real interface elements often beat pure chat.

Adoption, limits and early tests

TypeSafe says developers send Jev application data and typed questions, and it returns a pick from up to 255 options, a score, or a yes-or-no answer, each with a probability attached. It does not generate text. OpenAI, Amazon, Cloudflare, Perplexity and Microsoft have shipped decision models too. TypeSafe launched Jev on Sept. 15 with a $40 million seed round led by DCVC. On Friday, it said it raised an $870 million Series A led by Andreessen Horowitz at a $7.5 billion valuation. TypeSafe says a third of the Fortune 500 use Jev but has not named them; Andreessen Horowitz's announcement puts the figure at 25 percent.

TypeSafe's documentation says Jev struggles with counting, arithmetic and dates, leans toward whichever option is listed first, and does not treat injected content as hostile by default. In an OpenRouter Banking77 test, Jev answered in a median 175 milliseconds with 81 percent accuracy. Claude Opus 5 scored 84.4 percent but took 2.3 seconds and cost about 22 times as much. Andrew King of Very Good Ventures wrote that closed questions sent to Jev cut his generative dashboard's median render time to 3.8 seconds from 29 seconds with a frontier model on synthetic data. Anything under 50 percent confidence goes to a language model, a setup he called "a cage with a door."

Data handling and other paths to Sana

Workday recently added TypeSafe to its subprocessors list. The entry covers all Workday product lines, not just Sana, for natural language processing, intent classification and confidence scoring. TypeSafe says Jev runs only on its hosted API, processes data in the U.S., and offers zero data retention to enterprise customers. King wrote in September that he kept client data off Jev because the vendor was "weeks old with no published security certification yet." TypeSafe now offers a SOC 2 Type II report on request. Hellermark also showed Sana inside Anthropic's Claude. Workday's president of product and technology, Gerrit Kazmaier, called that an AI-to-AI integration, and said users lose the Jev-driven interface and some context.

Sources: The New Stack