The claim that AGI arrives by 2030 is a bet on the current paradigm continuing to scale while the missing pieces, continual learning and reliable autonomy, get solved as engineering rather than as research. It is a defensible bet. It is not the consensus.
What would have to be true
Three things. Capability would need to keep improving without a new architecture. Continual learning would need to work at production reliability. And the cost of running such a system would need to fall far enough to deploy it widely. Any one of those failing pushes the date out.
The counterargument
Every previous prediction of a near-term general intelligence, from the 1960s onward, underestimated the difficulty of the parts that are not visible in demos. Reliability under load is the classic trap.
How to hold the question
Treat 2030 as one scenario among several and build for the range. The organisations that handle the transition well will be the ones whose data, processes and people are ready regardless of the exact year.