Dictation for developers: talking to your editor without losing precision
Developers are the hardest audience for dictation. Code is dense with tokens no language model of English expects: camelCase identifiers, snake_case fields, file paths, flags. Say “user I D” to a generic transcriber and you get “user I.D.” — three characters of wrongness that break the build.
Tokens, not words
Developer Mode treats the editor as a first-class destination. It recognizes identifier intent and emits userId, user_id, or user-id based on the convention already around your cursor. Symbols come out as symbols: “arrow function returning null” is code, not a sentence about archery.
Where voice actually wins
Nobody dictates a for-loop character by character. Voice wins at the prose embedded in engineering: commit messages, PR descriptions, code review comments, docstrings, README sections, issue reports. That's a third of a working day for most engineers, and it's all text you can speak faster than you type.
The test is round-trip cost: if dictating a commit message takes one attempt and zero edits, you'll use it every day. If it takes a cleanup pass, you'll quietly go back to typing.