Wordware

Wordware is a San Francisco-based company that describes itself as an "AI context lab." Its original product, referred to on its site as Wordware V1, was an integrated development environment for building AI agents and applications using natural language rather than code. The company's thesis was that "words are the next programming language" — that prompting is a form of programming and that domain experts, not only engineers, should be able to specify and iterate on large language model behavior. Applications built on the platform were called "WordApps," and cited use cases included legal contract generation, marketing content automation, invoice analysis, candidate screening and PRD generation. The platform offered prebuilt tools, API deployment and tracing.

In 2025 the company pivoted from developer infrastructure to an end-user product. According to its own account, it launched Triggers & Actions with 2,000+ integrations in April 2025, observed that users built workflows but did not sustain usage, and concluded through mid-2025 that users wanted completed work rather than workflow construction. The company validated the direction with its lead investors, Nabeel Hyatt at Spark Capital and Wesley Chan at Felicis, in July 2025, and announced the pivot internally, moving from workflow builder to AI companion and from infrastructure to a prosumer product, accompanied by an architectural rebuild and team realignment.

The resulting product, Sauna, is positioned as an AI workspace assistant for knowledge work, described by the company as "Cursor for Knowledge Work." It is built around persistent memory and learning: user data such as emails, transcripts and documents is stored in a queryable file system; "Spaces" organize work areas with their own context and rules; "Recipes" cache repeated automations, including proactive ones that run in the background; and "Preferences" encode user style, priorities and boundaries. With user permission it connects to tools such as Gmail, Calendar and Slack. The design follows a programming-by-demonstration pattern in which the assistant observes a task performed two or three times, then offers to automate it, with human review and approval of outputs. Y Combinator's listing frames Sauna as an agent that earns autonomy task by task while the user retains review and final say.