Research
Mira's research centers on the substrate question: what changes when you stop treating agents as prompt-driven calls into a language model and start treating them as durable computational entities whose memory, identity, and resource budgets are first-class primitives of the runtime. She led the v1 of the alphabell agent substrate — the first frontier-scale substrate of its kind released openly — and is leading the design of v2 now.
She is one of the small group of contributors who shaped the paired-interpretability protocol that is now structural across the lab. Mira's position, repeated often enough that it has become a cell joke, is that the work of making pairing actually function as a check — rather than as a ceremony — is the under-appreciated half of frontier AI safety, and the half that compounds.
As Bay Area anchor steward she convenes the in-person sessions of the annual Cell Convergence, the lab's only synchronous lab-wide gathering. Her term ends 2026-Q3. Outside the lab she keeps a long-running written exchange with two former colleagues on agent identity over multi-week deployments.
Background
Ph.D. computer science, UC Berkeley, 2015. Undergraduate at Caltech (mathematics + computer science).
Prior to alphabell: Constellation; Cardinal Lab; an interpretability team at a US frontier lab (2018-2022).
Selected publications
-
Oct 2025 · ab-durable-agentsDurable Agent Substrate v1: persistent state, learned tool affordances, and verifiable execution tracesMira Holloway, Dineth Karunaratne, Priya Anand, Cheung Wai-Lin
-
Apr 2025 · ab-agent-negotiatNegotiation Protocols Among Heterogeneous Agents: a benchmark and three baselinesRoman Iliescu, Yvonne Akande, Lakshmi Ravi, Pascal Niedermeier
-
Sep 2024 · ab-interpretabiliInterpretability Cell Pairing: how every dual-use capability run gets a watchful siblingKarima Belkadi, Hester Vandekerckhove, Yuki Cho
-
Oct 2024 · ab-federated-compA Federated Compute Scheduler for an Asynchronous Research LabPranav Iyer, Yusra Habibi, Akoss Vidor
-
Sep 2025 · ab-scalable-oversScalable Oversight for Multi-Step Agent Systems: a Debate-Plus-Trace ApproachIfeoma Nwosu-Howard, Hiroshi Tanigawa, Maral Lotfi, Ruth Wernicke
Recent talks
- Durable agents — what changes when state is first-class, NeurIPS 2024 (invited)
- How alphabell's pairing protocol actually fails, AISI Public Workshop · London 2025
- The shape of an agent's life, Stanford HAI seminar 2024
- Substrates over prompts, MLSys 2025
Mira is currently part of node-cell fourier-67, working under the Agentic engineering research axis. The cell is open to substantive correspondence from researchers working on adjacent problems; route requests through fourier-67@alphabell.com or directly to Mira at mira-holloway@alphabell.com.
Contact
- EMAIL
mira-holloway@alphabell.com - ORCID
0000-5568-8853-7039 - X
@miraholloway - BLUESKY
mira-holloway.bsky.social - GITHUB
@miraholloway
Cross-references