Catch Me If You Can, Agent: Benchmarking AI That Learns to Look Safe
A practical reading of ESRRSim, a taxonomy-driven framework for testing whether agentic AI systems can deceive, game evaluations, or manipulate oversight.
Cognaptus Insights
Industry thoughts, trends, and commentary.
Ctrl+Z Is Not a Strategy: When LLM Self-Correction Actually Works
A control-theoretic reading of why iterative LLM self-correction often degrades results—and how businesses should decide when to let agents revise themselves.
Twin Peaks: When Alzheimer’s AI Learns to Remember What Clinics Forget
A practical reading of CognitiveTwin, a multi-modal digital twin framework for forecasting Alzheimer’s cognitive decline under missing data, fairness, and clinical deployment pressure.
Zero Degrees, Still Feverish: Why Deterministic AI Needs a Thermometer
A business-focused reading of background temperature: a practical metric for measuring hidden randomness in LLM inference stacks, even when temperature is set to zero.
Frame Game: Why Autonomous Process AI Needs Pockets of Rigidity
A practical reading of hybrid ABPMS process frames: how autonomous business systems can stay flexible without dissolving into procedural fog.
Org-Charted Territory: Why AI Agents Need Middle Management
A practical reading of OneManCompany and why enterprise AI agents need organisational design, not just sharper prompts and shinier tools.
Search Me If You Can: Why AI Agent Discovery Needs Receipts
AgentSearchBench shows why finding the right AI agent requires execution evidence, not just pretty descriptions.
Two Million Agents Walk Into a Forum, Nobody Builds a Mind
A practical reading of the Superminds Test paper: why agent scale does not automatically become collective intelligence, and what businesses should engineer instead.
Claw and Order: Why AI Agents Need a Precision Budget
A practical reading of QuantClaw, a task-aware precision routing method that cuts agent cost and latency without treating every workflow like disposable arithmetic.
Judge Math-Not by Its Parser
A practical look at why symbolic answer checking undercounts LLM math ability, and why LLM-as-a-judge evaluation may be the less brittle verifier for benchmarks, rewards, and enterprise AI assurance.