Iris ten Teije

Iris ten Teije is the co-founder of Sky Valley Ambient Computing and Differ, an infrastructure product for software that adapts to individual users after deployment. She is tackling a central problem in AI-generated software: how to let applications change continuously without sacrificing isolation, reliability, or accountability.

From fintech founder to adaptive software

Ten Teije studied at Amsterdam University College and became general manager for Europe at Neat, a financial-technology company serving entrepreneurs and small businesses. Working in digital banking shaped her appreciation for production environments where software failures carry immediate consequences.

She subsequently co-founded Koia, a platform for fractional ownership of collectibles, and advocated greater transparency about money and investing to expand women’s financial participation. She later co-founded and led Mava, which used AI to help organizations manage customer support across community channels including Discord, Telegram, and the web. Her writing on community-based customer support argued that public, conversational communities require different operating models from conventional support-ticket systems.

When Noosa Labs acquired Mava in 2025, ten Teije stepped away and co-founded Sky Valley Ambient Computing with Noam Tenne. Differ initially explored tools for people creating software with AI, then shifted toward helping existing applications adapt to their users after interviews suggested its original approach addressed less urgent problems. Her account of that early product evolution captures the strategic change.

The infrastructure behind individualized software

Ten Teije argues that conventional deployment pipelines distribute one fixed application because producing trustworthy changes was historically expensive. As AI lowers that cost, applications can evolve during a live session—provided their architecture preserves the safeguards once supplied by a frozen release. Her AI Engineer World’s Fair talk develops several defining ideas:

• A shared stem with bounded divergences: Each user’s application begins with a canonical version and develops isolated, reversible changes. Developers define what can adapt while excluding sensitive systems such as authentication and payments.

• Personalization beyond predefined segments: An investor using a CRM could automatically receive founder-introduction fields, reordered opportunities, and fewer irrelevant prompts without developers maintaining a separate application for every customer type.

• Coordination through shared intent: Ten Teije’s Intent Space approach organizes autonomous agents around declared goals, allowing individualized software versions to pursue common outcomes without requiring identical code.

• Provenance, validation, and measurable outcomes: Teams must understand why each change occurred, inspect what an individual customer is running, validate correctness, reverse failures, and measure effects on retention or support demand. For ten Teije, generating code is the accessible part; making autonomous customization trustworthy is the harder infrastructure problem.