One thing I've been noticing is that as agents get better at producing code, the bottleneck shifts even more toward maintaining architectural consistency across everything they're generating.
Getting an agent to build a feature is becoming relatively easy. Making sure ten different agent-generated features still feel like they belong to the same system six months later is much harder.
That's why the governance section resonated with me. In a lot of discussions, people focus on agent autonomy, but the real challenge seems to be preserving system coherence as throughput increases. Otherwise teams end up shipping faster while gradually accumulating architectural drift.
The interesting question for me isn't whether agents can write production code anymore. It's how teams prevent agentic velocity from turning into long-term maintenance complexity.