We've had a pretty wild new AI Agent feature that we've almost released twice, but then right as we get close, a new technical breakthrough makes the agent substantially better than before. The pace of new best practices emerging in AI is unlike anything that's happened in tech before. The common trend is that we're all collectively figuring out the best ways to split work up between agents to enable them to do much more complicated tasks, handling vastly more information than the context window can natively support, finding the most relevant data to work with on the task, and mitigating the natural limits on the AI models. What's amazing is that for many of these areas of improvements, you can simply decide to throw more compute at the problem at any point to unlock another tranche of performance improvement. Of course the ultimate constraint then is what someone will pay to perform the task, but it's amazing that we always have this variable at our disposal.