2025 - 2026 CE – The Agentic TurnBetween 2025 and 2026 the question stopped being whether a machine could answer and became whether it could act. Systems moved from producing text to holding a goal, breaking it into steps, running them, and correcting themselves along the way — the difference between an oracle and a worker.Two markers. Reasoning systems began passing the average human score on ARC-AGI-2, a test built specifically to resist memorisation and force abstraction from first principles. And agents crossed from demonstration into production: Gartner projects 40% of enterprise applications running task-specific agents by the end of 2026, up from under 5% in 2025.Notice what this does to the e/acc versus h/acc argument made above. While intelligence was something you queried, the dispute was philosophical. Once it is something that acts on your behalf, it becomes a question of whose behalf — and that is the whole of the argument.The purists have quietly retreated. 'AGI' as a single threshold moment is receding into the 2030s, because the scaling of language models is meeting physical and mathematical limits. What arrived instead is functional: agents doing long-horizon work across knowledge domains without a human in the loop. Whether that counts as 'real' AGI is a definitional argument. Knowledge work is being dismantled either way.