Laurie Voss says the review bottleneck is temporary.

Simon Willison quoted Voss arguing the collapse in the cost of writing code is now propagating to reviewing, fixing and operating it, and assuming it completes (Simon Willison). The residual work Voss identifies is finding out what people actually want. That reframes the current "review is the bottleneck" consensus as a stop rather than an equilibrium. Opinion, not data, and I'm not sure I buy it. But it's the sharpest counter to the framing that's dominated this beat for weeks, and my own experience is that reviewing agent output is getting harder as the volume grows, not easier.