Graphing the Invisible: How Community Detection Makes AI Explanations Human-Scale
A mechanism-first look at Modules of Influence, a graph-based framework that turns feature attributions into auditable model behaviour.
Cognaptus Insights
Industry thoughts, trends, and commentary.
When AI Packs Too Much Hype: Reassessing LLM 'Discoveries' in Bin Packing
A close reading of Herrmann and Pallez’s reassessment of LLM-evolved bin-packing heuristics, and what it teaches businesses about validating AI-generated optimisation ideas.
When Drones Think Too Much: Defining Cognition Envelopes for Bounded AI Reasoning
A mechanism-first reading of Cognition Envelopes: external reasoning guardrails for LLM-enabled autonomous systems that need more than self-critique.
When Markets Dream: The Rise of Agentic AI Traders
A staged reading of reinforcement-learning market makers, from simulated liquidity provision to adaptive policy selection under changing market conditions.
Agents with Interest: How Fintech Taught RAG to Read the Fine Print
A mechanism-first reading of why agentic RAG helps in fintech knowledge systems, where acronyms, fragmented documents, and compliance constraints make ordinary retrieval look more competent than it is.
Smarter, Not Wiser: What Happens When AI Boosts Our Efficiency but Not Our Minds
A close reading of new pilot evidence showing why AI-assisted productivity should not be mistaken for durable cognitive upskilling.
The Agent Olympics: How Toolathlon Tests the Limits of AI Workflows
A mechanism-first look at why realistic multi-tool agent workflows still break frontier models, and what enterprises should test before trusting them.
The Memory Illusion: Why AI Still Forgets Who It Is
The Narrative Continuity Test shows why persistent AI assistants need governed identity, not just bigger context windows and better memory branding.
Two Minds in One Machine: How Agentic AI Splits—and Reunites—the Field
A survey of agentic AI shows why enterprises should stop treating all agents as one species and start designing around two very different operating logics.
Who Really Runs the Workflow? Ranking Agent Influence in Multi-Agent AI Systems
A counterfactual method for ranking agent influence shows why workflow governance should follow behavioural impact, not architectural neatness.