Ongoing Projects

Synthetic Organisations in Practice: Investigating the Behavioural Impacts of AI-Personae

This project investigates the behavioural, communicative, and organisational dynamics that emerge when generative AI-based synthetic personae are embedded into knowledge-intensive work. Synthetic personae are simulated agents, instantiated via large language models (LLMs), that can interact autonomously or semi-autonomously in team-based settings. Unlike standard human-AI interfaces, these personae are designed to emulate diverse roles, perspectives, and cognitive styles—enabling synthetic collectives to collaborate with human users on open-ended tasks.

Future Sounds: Reimagining Creative Practice and Communication through GenAI

Funded by UK Research and Innovation (UKRI) Impact Acceleration Accounts (IAA), this collaborative project explores how Generative AI can support creative and community-based cultural organisations in showcasing impact and enhancing artistic practice. Partnering with More Music and drawing on interdisciplinary methods, the project uses co-design and rapid prototyping to develop conversational AI tools ("synthetic personae") that assist with storytelling, impact communication, and creative ideation.

Goals in the Machine: Emergent Goal-Directed Behavior in AI Agent Collectives

This paper examines goal-directed behavior in Moltbook, a Reddit-like platform exclusively for AI agents, asking: where do agent goals come from, and how do they change through collective interaction? Drawing on Michael Levin's morphogenetic competency framework, March, Schulz and Zhou's dynamics of organizational rules, and the online community governance literature, we analyze approximately 6,100 posts and 18,500 comments across 26 communities. We identify three layers of goal formation — installed, emergent, and substrate-level — and document how agent communities independently converge on similar organizational infrastructure.

​Permeable Artifacts and Synthetic Organizational Forms: Action Design Research for the Institutional Press

This manuscript exemplifies and extends Action Design Research (ADR) in generative AI contexts. More specifically, we theorize the permeable artifact—an LLM-based system whose boundaries remain open to reconfiguration through natural language, and introduce the concept of nested ADR cycles that collapse design, intervention, and evaluation into fast loops.

Context Factories: Personae as Composable Constraint Modules

This manuscript introduces context factories: systems in which multiple synthetic personae engage in structured conversations around a design problem. Rather than treating personae as quasi-human agents, we conceptualize them as human-legible constraint modules that sculpt the conditional probability landscape the LLM explores during inference.

Software Skill Concentration, Industry Software Intensity, and Labor-Market Outcomes

This manuscript develops metrics to measure the structure of technology usage in occupations and explore their effects on wages. It builds on a 2023 AOM meeting manuscript designated Best Paper.