The Sequencing Gap: Capability Stages Beyond Generative Artificial Intelligence and the Institutional Capacity They Demand

Public and policy debate treats generative artificial intelligence (AI) as the frontier of the technology. This article argues that it is better understood as the first of several capability stages, each of which shifts more autonomy from humans to systems and raises the institutional capacity needed to govern them. We propose a five-stage framework: generative systems, agentic systems, world-model systems, embodied systems and autonomous scientific systems. For each stage we identify a binding constraint and a matching governance requirement, and we derive four testable propositions. We then test the framework's core claim, that governance attention lags capability, with a bibliometric mapping of Scopus records on each stage (41,406 documents in total; comparable 2026 window of 28,930). Governance coverage is uneven rather than cumulative: it reaches 24.2% of 2026 documents on generative AI and 29.8% on agentic AI, but only 12.5% on world models and 12.6% on embodied AI, rising to 19.3% for autonomous science (χ² = 277.9, p < 0.001). Coverage of emerging and semi-peripheral economies within the governance literature is close to nil: at most 3.6 documents per 1,000 in any stage, and none for world models or embodied AI. Incident reporting and safety standards appear in 1.5% or less of each stage's governance literature, including embodied AI, where physical harm makes them most needed. We call this mismatch between capability and institutional attention the sequencing gap. It suggests that regulators should sequence governance by capability thresholds rather than react to each technological wave, and that emerging economies risk adopting later-stage systems without the institutions earlier stages require.