Resources

What we read.

Source material for the supervisory thesis. Papers we cite, regulatory texts we map to, reports that frame the market.

The $3 Billion Productivity Gap: Why Small-Medium Businesses Operate at 47% Efficiency

The structural inefficiency that supervision-by-architecture absorbs. McKinsey on the 47% productivity gap and the SaaS-automation pattern in regulated back-office work.

“Small businesses are 47% as productive as large firms, and this productivity gap is equivalent to 5.4% of US GDP… Technology adoption through SaaS providers automating back-end operations like compliance and accounting has saved businesses significant time.”

— via McKinsey Global Institute | Small Business in America: Time to think big

Self-Evolving Multi-Agent Systems: Agentic Neural Networks for Automated Decision Processes

Coordination as architecture. The paper’s neuro-symbolic baseline informs how we bound agent action spaces in production — specialist teams reorganizing without re-engineering, but inside a permission envelope the architecture enforces.

“Across four benchmark datasets, ANN surpasses leading multi-agent baselines under the same configurations, showing consistent performance improvements… This neuro-symbolic approach enables ANN to create new or specialized agent teams post-training, delivering notable gains in accuracy and adaptability.”

— via arXiv:2506.09046 | Ma, X. et al. “Agentic Neural Networks: Self-Evolving Multi-Agent Systems”

Beyond Chatbots: Multi-Agent Chat Systems That Generate Domain-Specific Business Intelligence

Bounded specialist agents collaborating via template-driven workflows. The Propose phase of the supervision loop, in research form — context-aware role allocation rather than a single unbounded chatbot reaching for everything at once.

“Chat-of-Thought employs multiple collaborative Large Language Model (LLM)-based agents with specific roles, leveraging advanced AI techniques and dynamic task routing… This research demonstrates the potential of Chat-of-Thought in addressing these challenges through interactive, template-driven workflows and context-aware agent collaboration.”

— via arXiv:2506.10086 | Constantinides, C. et al. “Chat-of-Thought: Collaborative Multi-Agent System”

How to Count AIs: Individuation and Liability for AI Agents

The legal individuation problem and its engineering answer. The article’s “thin” identity (every AI action tied to a human principal) is exactly what bounded action spaces and bar-number-attributed audit trails deliver at the architectural layer.

“Two kinds of identity are required: “thin” and “thick.” Thin identification ties every AI action to some human principal, essential for holding accountable the humans who make and use AI agents. Thick identification distinguishes between AI agents, qua agents — sorting millions of AI entities into discrete, persistent units with stable, coherent goals.”

— via arXiv:2603.10028 | Arbel, Salib, Goldstein. “How to Count AIs: Individuation and Liability for AI Agents”

Reading something we should be reading? Send it.