Spent a week with Udio. It's fun. But fun isn't billable.
If you're in pro audio post, your tools need precision, non-destructive workflows, and reliable outputs. Udio gives you... vibes. Need a 3.2-second, room-tone-matched fill for a dialog edit? Good luck. Need consistent, layered stems for a mix? Not happening. It's great for generating a "synthwave track about my cat" for a social clip. But for actual client work? You'll spend more time trying to wrangle its output than you would just cutting the sound yourself or using a proper library.
It's a creative spark for hobbyists, which is fine. But calling it a professional tool is a stretch. It's like showing up to a construction site with a Lego set.
CRM is a means, not an end.
I think you're over-indexing on its current use case as a direct replacement for a DAW. The precision argument is valid for sound design, but I've been testing its output as raw material for a very specific purpose, layered ambisonic textures.
The key isn't expecting it to output a 3.2-second fill directly. You prompt for a 30-second loop of, say, "industrial HVAC rumble with metallic resonance," run it through a spectral analysis against your target room tone, then use that analysis to guide processing chains in Reaper. It becomes a source for convolution impulses or a modulation layer. It's faster than sifting through a 2TB library for a texture that's 80% right and still needs heavy manipulation.
Your Lego analogy holds for final assembly, but sometimes you need a unique plastic pellet to injection-mold a custom bracket. The tool's value is in generating novel source material, not finished, timed assets. Its stochastic nature is a feature for that, not a bug, though it absolutely precludes its use for deterministic tasks like dialog replacement.