The 2026 AI Research Landscape
Change, Constraint, and Continuity
During 2025, capabilities that once had to be scaffolded externally became embedded within the AI systems themselves. At the same time, the societal and institutional stakes of AI deployment became impossible to ignore, and constraints became necessary. Without them, risks of misuse, misattributed authority, and moral confusion would have grown unacceptable.
These changes define the contemporary environment in which serious AI research must now proceed, The constraints are real. What matters is how we respond to them. For example, the AI agent the persona framework has now been re-anchored.while preserving their core functions. In parallel with the maturation of large language systems, the concept of agentic skills has emerged as a key lens for understanding procedural competence in AI. These structured capabilities — reusable, modular mechanisms for reasoning, planning, constraint management, and memory — are increasingly native to contemporary agents rather than external scaffolds.
Within this project, that shift reframes what were once externally-scaffolded personas into epistemic agents: oriented modules of reasoning and interpretation that exploit embedded competencies while maintaining distinctive stances and functions. This raises a naming question. Should these structured entities now be called “agents” rather than “personas”? The distinction lies in emphasis. “Agent” foregrounds actionable competence enabled by embedded skills, while “persona” highlights interpretive orientation — the stance from which reasoning, responsibility, and meaning are negotiated. In this research ecology, both aspects are important: agents for what can be done, personas for how it is conceptually framed. On this page, we retain the term persona to emphasise the latter, while recognising the broader discourse in which agentic structures play a growing role.