Does This Task Need an AI Agent? Two Decisions, Not OneChoose between rules, LLM workflows and agents, then separately decide which actions the system may execute using evidence, permissions and outcome checks.
How Much Uncertainty Can an AI Agent Act On?Define the boundary for autonomous action: meaningful falsification or a validated dynamic heuristic that detects failure before unacceptable consequences.
The Same Task, Different Autonomy: Research, Draft, PublishA UniMatrixZ target pattern showing bounded research, reviewed drafts and controlled publication, with explicit acceptance gates and verified outcomes.
The Trojan Horse of Free Frontier AIWhen free AI becomes scientific infrastructure, its capabilities can shape which questions get asked—and which alternatives remain worth exploring.
AI Transparency Is a System PropertyA model explanation is not enough. AI transparency must be designed into the surrounding knowledge and decision system so outputs can be traced, challenged and corrected.
Falsification, Not Determinism, Is the Central AI ProblemDeterminism supports reproducibility, but reproducibility is not truth. Knowledge-centric AI requires claims that can be tested, falsified and corrected.
