← All speakers

Bio, Work & Ideas

Tereza Tížková

Conference affiliation: Growth · Factory · 2026

Tereza Tížková’s 2026 AI Engineer conference biography describes her work in growth at Factory, an AI-native software development platform whose autonomous coding agents, called Droids, help engineering teams ship across the development lifecycle. Before Factory, she worked on growth as the first hire at E2B, which provides open-source sandboxed virtual machines powering AI products. Outside work, she enjoys blogging, testing new products and talking to developers.

In her presentation on software factories, Tížková describes an autonomous lifecycle encompassing user feedback and logs, prioritization, execution, validation and continuous improvement. She explains Factory’s automatic model routing, which classifies task difficulty and selects the cheapest model predicted to meet the required capability threshold. She also presents Factory Missions: an orchestrator establishes completion criteria, sequential workers hand off work with fresh context, and independent validators assess code against a contract written before implementation. These checks include an agent that clicks through an application on a virtual computer. Her discussion of continuous learning covers deferred context that loads full tool details only when needed, agent-readiness checks for reproducible environments, tests and documentation, and plugins that package reusable skills and team knowledge. She argues that humans should focus on deciding what software to build while agents handle implementation and routine coordination.

1 conference talk

Key ideas

Scroll to read ↓

Tereza Tížková explains how Factory connects model routing, sequential workers, independent validation and selective context loading into an autonomous software lifecycle—and why codebase hygiene and human judgment still matter.

  • A software factory connects signals, prioritization, implementation, validation and learning into an autonomous cycle.
    1:27 ↗
  • Model routing depends on predicting task difficulty well enough to choose the cheapest capable model without frequent switching.
    8:12 ↗
  • Completion criteria must describe the intended result. Independent code checks and application interaction help expose failures that merely passing tests can miss.
    11:45 ↗
  • Deferred context keeps full tool specifications available while loading them only when needed; codebase hygiene and reusable team knowledge support the quality of subsequent work.
    16:51 ↗
  • Tížková envisions humans choosing what to build and monitoring agent teams, with agents also taking on routine status sharing and coordination.
    21:25 ↗

References