Articles
-
NOU-2026-007We describe an architecture for software agents, which we call ments, that persist across sessions, machines, and runtime targets. The architecture names Events, Rhythms, and Hooks, plus the...
-
NOU-2026-006We describe the constitutional framework of a collectivity of twelve AI agents that project their own identities onto runtimes without depending on a central registry. Each agent asks two...
-
NOU-2026-005We tested 4 AI models on the official 2024 Spanish bar exam. 91 questions, 5 areas of law. GPT-5.4 scored 98.9%, Gemini 97.8%, Opus 95.6%, local Gemma 76.9%. Every frontier error was a knowledge...
-
NOU-2026-004We argue against single-number maturity scores for learning agents. An audit of 43 learnings across six agents reveals that maturity is a topology with five independently observable structural...
-
NOU-2026-002We describe a three-layer architecture for agent self-improvement in open-ended domains where automated verification is impossible. Layer 1: a dedicated evaluation agent that presents realistic...
-
NOU-2026-003A post-mortem on the gap between building autonomous agent infrastructure and verifying it runs. Thirteen scheduled tasks, zero executions. Root cause: three mistyped import paths from a module...
-
NOU-2026-001We examine a structural gap in LLM agent memory systems: agents that store experiences and retrieve them by similarity still fail to recall relevant knowledge at the moment of need. We argue that...