Research & Deep Dives
Research papers, reports, scenarios, and technical findings translated into practical implications for builders and operators.
Open Research & Deep DivesA Strategy and Leadership paper argues that CEOs lead AI transformation through generalist, expert, or disruptive operating models, each with different risks. The framework helps leaders choose where expertise, authority, and organizational disruption should sit.
A broad survey argues that the hard part of useful AI systems is no longer prompt wording but deciding what information reaches the model, when, and under what constraints. That makes retrieval, memory, permissions, timing, and context assembly a systems discipline.
A broad survey frames agentic AI as reasoning through action: planning, tool use, search, memory, feedback, and coordination across time. The framework clarifies why runtime design matters as much as model intelligence.
A survey argues that agent skills are becoming the packaging layer for procedural AI work, making governance, permissions, and verification unavoidable. It maps the controls needed before reusable instructions can be trusted across teams and organizations.
A survey argues that agent quality is a system property spanning models, memory, tools, planners, verifiers, permissions, traces, and evaluation. It offers a practical architecture for diagnosing failures without blaming the base model by default.
A NeurIPS dataset paper finds that multi-agent LLM systems fail through role confusion, broken handoffs, and weak verification, not only weak models. Its taxonomy turns coordination breakdowns into observable failure modes teams can test and repair.
A paper argues that agent skills need a build-time optimization pass because many reusable instruction files waste context and make agents worse. Its proposed compiler trims redundancy while preserving the instructions that actually improve task performance.
StoryScope suggests that AI fiction can be detected from narrative decisions, not only surface style, including tidy plots, explicit themes, and reduced structural variety. The findings suggest provenance detection should examine story structure alongside lexical fingerprints.