Comment by stopyellingatme

Just as an anecdote, i used it to review a PR with 24 file change. We pivoted from the initial draft to make a service bus subscription more lightweight and use SignalR to update the frontend.

It used 1.4 million tokens and 34 sub agents during its review. This was not a large PR. So my read is that its very thorough, not good to use it for "small/medium" tasks unless precision is a very high priority.