The Tail That Wags the Model: Why p99 Latency Should Run Your LLM
A practical reading of SLO-Tuner: why LLM serving teams should optimize p99-satisfying goodput, not average latency, raw throughput, or speculative decoding bravado.
Cognaptus Insights
Industry thoughts, trends, and commentary.
From Durations to Dynamics: Translating Temporal Planning into PDDL+
A mechanism-first reading of how temporal numeric planning can be compiled into discrete PDDL+ without quietly breaking the semantics that make schedules valid.
Green Lights, Smarter Cities: How Multi‑Agent Reinforcement Learning Is Rewiring Urban Traffic
A mechanism-first reading of how robust training, driver-compatible signal actions, and neighbor-level coordination make MARL traffic control more deployment-ready.
Print Smarter, Not Harder: How Portfolio Algorithms Are Quietly Optimizing 3D Printing
A mechanism-first look at why Portfolio-CEGAR-SEQ improves sequential 3D printing by running diverse packing and ordering strategies in parallel rather than betting on one clever heuristic.
Too Smart to Share: When AI Agents Get Smarter, Systems Get Worse
A mechanism-first reading of why more adaptive AI agents can overload shared resources under scarcity—and why capacity per agent should be checked before upgrading intelligence.
Topology Trouble: Why Even Frontier LLMs Still Get Lost in a Grid
TopoBench shows that many LLM failures in spatial reasoning come from weak constraint extraction, not merely weak reasoning.
Agents With Memory: Turning Execution Logs into Institutional Knowledge
A mechanism-first reading of trajectory-informed agent memory, showing how execution logs can become structured operational guidance rather than decorative vector-store clutter.
Audit the Bots: When AI Judges the Work of Other AI
A practical reading of CUAAudit and what its evidence says about using vision-language models to audit autonomous computer-use agents.
Diagnosis, But Make It Iterative: When AI Learns Like a Doctor
DxEvolve shows why governed clinical AI may depend less on bigger models and more on workflow-constrained evidence acquisition plus auditable experience memory.
Don’t Build the Agent — Raise It: The Nurture‑First Paradigm for AI Expertise
A mechanism-first reading of Nurture-First Development, a framework for turning practitioner-agent conversations into reusable domain expertise.