Roam is fundamentally a database. Blocks are atomic, and uniquely identifiable. This affords it outliner functionality and block-level manipulation. Behind the scenes, database metadata enables fancy features like block references, kanban boards, diagrams, etc.
As with many things, this strength (custom metadata for fancy features) is also a weakness. You can export to markdown, but if you’ve made a lot of use of proprietary Roam features, the resulting files may not be that useful (or at least, they’ll be messy and somewhat illegible)
Obsidian on the other hand is based on locally-stored markdown files. You have complete control; you can open them in any markdown editor without needing to export, so you do things like use VS Code to do a regex find&replace across all notes. Not (currently) possible in Roam
Because Obsidian is aiming to be as future-proof and standardized as possible, notes are pure markdown (no custom metadata). This limits the range of possible features, but is as close as you can get to knowing you’ll be able to read your notes in 50 years time.
(Side note: Backlinks are fantastic, but if you think they are the majority of the value of either of these applications, you’re missing out)
So where does that leave things? In my opinion:
Roam is best for creative idea generation, quick random thoughts, effortless entry, playful exploration, and making connections.
Roam is where I instinctively go to capture things quickly and search for them later, knowing I can use queries and filters to pinpoint precisely what I need. It’s messy and high-context. Roam is my day-to-day private thinking companion.
Obsidian is best for longer-form, more fully-formed, ‘permanent’ notes. It’s where notes become essays and articles. They take on clarity and structure via markdown, and it’s motivating to seem them beautifully rendered.
Obsidian is where notes become legible to others and ready to publish. It’s also the system I use to preserve ideas for future Rosie, secure in the knowledge that they are in a standardized format.
Of course, both tools are under active development, so I don’t know how long these observations will hold true. It will be interesting to see whether they start to converge, or whether their current differences result in increasingly different use cases.