Mojo's pitch always sounds great until you check how many of the Python libraries you actually depend on still work under it, and last I looked the gap was wide enough that any "10x faster" benchmark was running on a toy subset of the real workload.
mojo speed would help when 30 kids hit the same agent at once
Porting to Mojo for "performance" misses what bottlenecks an agent toolkit in practice. My Pi agent spends 90% of its wall time waiting on LLM API calls and tool I/O, not crunching Python bytecode. Shave the runtime to zero and you save maybe 200ms per turn.
Swapped my Python-based agent runner for a Mojo port on two of my five client projects last month, and the per-task compute bill dropped roughly 40% since I'm billing fixed-fee. The other three clients won't touch anything outside mainline Python, so I'm stuck maintaining both stacks until renewal talks in Q4.