Beyond Conversation: Why Documents Transform Natural Language into Code
AI Engineer World's Fair 2025 · 10:57
AI assistants and knowledge-work automation
Wordware builds Sauna, an AI workspace for individuals and teams that combines persistent context with background task execution. It uses emails, documents and other connected information to draft messages, prepare briefings and file tickets while users are offline. Available through web, iOS, Slack and other messaging channels, Sauna supports shared team memory and connections to more than 3,000 tools and apps, plus custom MCP connectors.
Cofounders Filip Kozera, CEO, and Robert Chandler, CTO, initially built Wordware as a natural-language development environment where engineers and domain experts could create AI applications together. Its infrastructure included prompt version control and production deployment. In 2025, the company pivoted toward Sauna, moving from tools for building workflows to an assistant that performs knowledge work. Its architecture stores context in a queryable file system, organizes work into spaces with their own rules, and captures recurring tasks as reusable automations. Users review, approve or edit the assistant’s work.
In 2024, Wordware raised $30 million in seed funding led by Spark Capital, with participation from Felicis and Y Combinator. At that time, the company reported hundreds of thousands of users and named Instacart, Runway and aiSDR as organizations using its AI development platform.
AI Engineer World's Fair 2025 · 10:57
AI Engineer World's Fair 2025 · 6:50
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here to understand why conversational context can become polluted and precise iteration difficult, motivating documents as editable specifications.
Filip KozeraAI Engineer World's Fair 2025
Follow a failed Slack messaging task and a demonstration combining Claude, Twitter data, and Notion to see how Chandler motivates and illustrates agentic tool design.
Robert ChandlerAI Engineer World's Fair 2025
Kozera's talk extends beyond document interfaces to agent communication: coordinating background workflows requires protocols that express constraints and authority beyond current MCP capabilities.
Chandler identifies excessive tool choices, verbose JSON, pagination, and authentication friction as sources of agent failure. He proposes natural-language interfaces to specialized tools as a way to reduce long reasoning chains.
Affiliations reflect each recorded session, not necessarily current employment.