Théo Fagnoni

Research

6 published papers, grouped by topic.

Coordination when humans and AI agents act on shared work: when an action may change shared state, which actions require review, and how the resulting updates reach every participant.

Théo Fagnoni, Phillip Kingston

A coordination framework for humans and agents on shared tasks: actions are admitted only if their inputs are current, permitted and approved. Sharing intentions cuts preparation attempts by up to 53.6%.

Mar 2026

Generating executable workflows from business intent: capturing intent from requests and documents, generating workflows as structured graphs, and checking that a workflow under construction can still be completed correctly.

Phillip Kingston, Théo Fagnoni

Formalises incremental workflow-graph construction in which every addition carries a certificate that a valid completion still exists, and characterises the coarsest abstraction that tracks this exactly.

ICMAI 2026 · Aug 2026

Théo Fagnoni, Mahsun Altin, Chia En Chung, Phillip Kingston, Alan Tuning, Dana O. Mohamed, Inès Adnani

Adds an intention-capture layer between a user query and workflow generation. On 1,000 synthetic multi-intent queries it improves semantic workflow similarity over direct generation, most for mixed intents.

Jul 2025

Phillip Kingston, Théo Fagnoni, Mahsun Altin

Defines Workflow Intention, the input, process and output that set a workflow's objective, and proposes a multimodal, attention-based encoder that infers it from business documents.

Feb 2025

Théo Fagnoni, Bellinda Mesbah, Mahsun Altin, Phillip Kingston

Generates executable workflows from client intent using a Large Work Model informed by a Work Knowledge Graph, then optimises them as graphs. Beats leading LLMs by 38% on medical coding.

Nov 2024

Quantifying workflow quality: expected performance in terms of success, cost and output, together with structural criteria, used to score, rank and optimise workflows.

Alan Seroul, Théo Fagnoni, Inès Adnani, Dana O. Mohamed, Phillip Kingston

A probabilistic-normative framework for scoring workflows: an expected reward over success, cost and output gain, plus penalties for cohesion, coupling, observability and information hygiene.

Nov 2025