Iben Lykke

Research

Iben leads ramanujan-07, the cell working on the tokenizer and compositional vocabulary layer for substrate-hosted agents. The cell's premise is that the way an agent's actions and observations are tokenized is a substrate-level concern, and that current tokenizers are not the right primitives for agents that need to compose tool uses over weeks.

She has been with alphabell since 2020 and is one of the contributors who works comfortably across NLP venues and AI-safety venues. The ramanujan-07 line of work has produced two cell-internal reports that are widely cited within the substrate work and that she expects to release publicly in 2026.

Iben is based in Copenhagen. She maintains a personal tooling library for tokenizer analysis that several other cells use.

Background

Ph.D. computer science, University of Copenhagen, 2015.

Prior to alphabell: Copenhagen; Wayfarer Institute; Hyperion Cognitive Lab.

Selected publications

• Oct 2025 · ab-durable-agents
  Durable Agent Substrate v1: persistent state, learned tool affordances, and verifiable execution traces
  Mira Holloway, Dineth Karunaratne, Priya Anand, Cheung Wai-Lin
• Apr 2025 · ab-agent-negotiat
  Negotiation Protocols Among Heterogeneous Agents: a benchmark and three baselines
  Roman Iliescu, Yvonne Akande, Lakshmi Ravi, Pascal Niedermeier
• Sep 2024 · ab-interpretabili
  Interpretability Cell Pairing: how every dual-use capability run gets a watchful sibling
  Karima Belkadi, Hester Vandekerckhove, Yuki Cho
• May 2025 · ab-mechanistic-ci
  Mechanistic Circuit Analysis at Frontier Scale: cells as a unit of interpretability
  Jiang Yifei, Nico Almgren, Karima Belkadi, Hester Vandekerckhove
• Sep 2025 · ab-scalable-overs
  Scalable Oversight for Multi-Step Agent Systems: a Debate-Plus-Trace Approach
  Ifeoma Nwosu-Howard, Hiroshi Tanigawa, Maral Lotfi, Ruth Wernicke

Full publications index →

Recent talks

• Tokenizers as a substrate-level concern, ACL 2025
• Compositional vocabularies for agents, EMNLP 2024