These days I see a bunch of discussions about sharding and parallel processing, and the narrative is quite lively, but my first reaction is: how exactly does the money get out? Is the exit path clear?

Many projects talk about throughput and modularization, but when it comes to actually withdrawing, the bridges have to queue, on-chain gets congested like a PPT, plus a bunch of permissions that haven't been reclaimed...
In short, no matter how much technical effort is put in, if the asset security line isn't maintained, it's all pointless.

And then there's the set of practices around pledge collateral and shared security, where the layered yields get called out as "copy-paste" schemes, which I can understand.
You wrap risks layer by layer, making it look more stable on the surface, but actually harder to dismantle underneath. If something goes wrong, it's unclear who will take responsibility.
Right now, I’m used to first checking permissions, then seeing if I can revoke authorization/unpledge with one click, and only then look at the yields.
It's a bit more trouble, but I sleep better.