Caveats would be.... that the "apply to Active over Selected" approach (from what I read also happen in Maya sometimes) is a huge one, a real limiting time sucking feature. Another is the polycount limits in the viewport and playback. I sculpt on meshes of up to 2-4 million polys and things can't get done eventually hitting memory leaks. I am about to buy more ram to see if that's the issue, but I find it's handling of geometry significantly less than SI. The animation is on par concerning the cool dopesheet, but I think the lack of Animation Layers nor the ability to "add" or "layer" animation clips to be a huge missing feature typical of Softimage and Maya.
For some procedural stuff, I've found a lot less flexibility to SI, but there is still some flexibility, layering modifiers, specially with addons like Animation Nodes or other addons that give a bit more flexiblity. I am no procedural guru, but I like the artistic approach of retopology with grease pencil strokes and easy IK/FK setups and the like overcome that - also the hair instancing for scattering, or other little thigns. Python scripting is really easy, and the API relatively quite well documented, though still strange - and I have to admit, the output console, compared to SI sucks. Operators are context sensitive so some repetitive python script operators are harder to code due to that fact. I still have troubles coding anything due to context issues because of it's super flexible interface.
But for file sizes, resources, and the quick and easy fur, the free cloud rendering with Sheepit, sculpt tools, sculpt animation, quick shape manager with GP and animation workflows, and quick rigging... Rigging is pretty easy, and doesn't crash when editing shapes, bones are pretty flexible and stable, performance does have opensubdiv thoughI haven't tried it, and there are crowd plugins being developed. I actually do my facial rigging without drivers and use Animation nodes entirely. There is a bit of a performance hit with those nodetrees, but it works. I have to run scripts to make keyingset and import/update the rigs in scenes to make sure the info bakes before render time, but that is a typical workflow of most cached animation to render farm workflow. Management of data is weird, like.. no EMDL like workflow, but with whole scene data blocks.. which has it's strengths, but also weaknesses with production nightmares when it comes to rendering in the cloud unless you pack everything in making large files to upload. But.. rendering is ok with Cycles on GPU, pretty straight forward, and beautiful, pretty quick, shading is very good and easy with the node editor (which also is really easy to use and has great node workflow hacks). Filmic Blender and other addons are a must, and the compositor is more than capable for most composition needs (just a little slow to render, but you can do that over the cloud and/or a farm). The motion tracking is the best one out there for free. The video editing system it has already is actually quite good, and.. what I use now for my video. I know a guy who edits 4K on Pentium 4's with Blender and it's proxy system (and even render out the 4K on the Pentium 4). The way you can use 3D scenes in the VSE also, with cameras, markers, grease pencil and it's strokes on geometry, all of that is stellar and I don't know of a better Animatic system out there. SI can do that, but you have to do some workarounds. The greasepencil is pretty straighforward and actually really really useful. I feel Blender is creative in that regard, concerning making sequences and the artistic workflow with sculpting with poly modeling (the modeling is alright, despite what they say, specially with some cool addons, for example the rounding script that came out of Maya 2018 just now was in Blender for years - but NO WAY compares to the dynamic operator modeling workflow of SI). Also, the ability to have MULTIPLE scenes in the same project is actually really awesome. Once you get the hang of the linking concept, it's really usefull for some productions and collaborative workflows (linking in, appending, making local, referencing, data management).. it's very flexible with that if you know what you are doing. Like I block characters in one scene, the animator has the blocked character linked in, and that blocked character updates as soon as the scene saves, uploads to the cloud and he updates and viola, the ghost blocked character updates (or any other linked data). We both can technically work on the same scene but working in linked scenes that won't conflict in versioning. I have yet to work hard on animation with it - but unless your animator is heavily dependant on clipping animation libraries and offset keyframe workflows, the dopesheet and graph editor are pretty easy and straightforward, and basic animation libraries can be built, all of it quick and fun to use with different key types, and clear visual language. Though playback is.. I keep SDS off, framerates suck. But 2.8 should fix that maybe, and I haven't tried Opensubdiv yet. Oh, and the weight painting.. is.. alright. A little more difficult than SI, but it's still pretty easy - just miss the smooth weights feature (but you do get healing, stepping, smooting individual ones, etc). Changing geometry with shapes is tricky though, but it doesn't crash things, though editing shapes is definately not an issue. It does have some ability to transfer weights and some properties from different meshes to others too.
I have done some particle work, instancing leaves with cloth attributes, and they did them all without much of a hitch (and only on modifier stacks) but it works. The solidify modifier is really good and fast. It's baking is pretty alright, controllable and nice with Cycles. Concerning features.. it has what you need. Concerning the UI, BFA has helped on some things but some things will always be Blender as much as Maya will always be Maya. There are many addons that work around some issues, like copy attributes to others, pie menus, etc. And they are useful.. but you have to know them.
Oh and.. objects are in world space, but parents seem to go to a weird local space. There is a Delta space, no idea how that works other than possibly 1 layer animation. So.. it's not as clear cut as SI.. and as a thread in the SI mailing list says, Maya has that issue too. But you can animate to transform or delta movement, and it will save both - but in Global space. But it's weird as the Global position now becomes a Local/Global position relative to the parent if you're animating a child. So you can have "Local" position if your object is a child - which is... "cool"... Render layers, with AOV's work, and with limited overriding - but no object/item override system as of yet, just shaders. Technically in 2.78 this will be possible with the Depsgraph, so I'm really looking forward to seeing how that works. The layer system is limited to 16 layers, but that's ok. Rarely have I needed more in SI. And with the linking/scene system, you can create your "partitions" almost to similar funciontality as Softimage. There aren't too many render engines integrated in, but Vray and Thea plug in. Cycles too. Alembic, less exact than SI concerning UV's I've found, but I hear they are improving that. The FBX workflow to and from different areas are about as easy and normal as elsewhere, including SI. It's animation baking is... once you understand the term, manageable. It's.. adequate all around concerning all areas. But personally the only thing I see it REALLY excells in is the Grease Pencil for 2D animation or collaborative work, or even quick artistic workflows.
Another comment on it would be.... ehm... it can get the job done. Maybe in more time than SI - but the creative preprodution phase, even with it's limited SSAO and camera DOF in the viewport is certainly great, in my opinion. And soon more so when it comes to using Eevee.
So all round.. it is nice having sculpting for free, despite the poly limit, and everything else for free. The UV workflow and unwrapping is ok - though I miss the limit loop select SI has, the poly modeling is alright. I have worked on some projects and it seems the largest studios here in the city use Blender, Unreal Engine, a mix of both - then Max and Maya. They made 8x10 minute episodes in a quasi 2D/3D animation with card rigs in 3D space in Blender over 6 months, had 60 average people on the team. Not sure if that is fast or not, but it worked. It was an older version too. Another studio used Blender for motion capture to produce 2 hours of content over a year and half.
But.. I can't say it's great. I just use it out of need and legality and to get work with it per requirement (so far so good).
I should probably get back to work..
Point form roadbloack deficits Go!
-Animation Offset Layers missing
-Animation Clips with Offset keyframe/Add keyframe workflow missing
-More integration and optimization of the node systems that already exist
-Ability to render out multiple scenes from the same file without Bat scripting it
-No clear Global/Local location system (Maya deficit also)
-Poly limits when sculpting/unwrapping
-Optimzed texture painting (goes slow at 4K)
-"Apply to Active over Selected" workflow everywhere (get rid of that in the Outliner or to Modifier editing and it will double it's useablility and efficiency)
But to be honest, I try stay away from the Blender community, and assess myself. But so far those who have used it whom I know have said that they prefer rigging in Blender out of Max, due to constraints variety, and others on big productions prefer Blender mainly due to the flexibility to script and find ways to script in Blender, also it's easier for new people to pick up the pipeline and learn than the software they already "know" like Maya or Max that they teach in Uni's here. Biggest complaints are in UI workflow, and noone really knows why.. but coming from Softimage, it's easy to see why. Same symptom of Maya UI improvements and the people who use it.