▸ T0 · main thread · write Substack post from build log · interactive
The comment engine runs outside the dashboard. That's the whole design.
Reply windows on X close in about 90 minutes. A comment that lands after that is noise; one that lands in the first 15 is signal. No human workflow is fast enough to catch every opening across three platforms, which means the automation has to be autonomous or it has to be something worse: manual and slow.
So the dashboard covers the parts that benefit from eyes before they go out. Drafts the pipeline generates automatically, ready to edit inline and publish now or schedule for LinkedIn, X, and Threads. A manual compose path for when I want to write something from scratch. All of it waits for a human decision because posts are slow: a LinkedIn essay or an X thread earns most of its engagement over hours, not seconds, so a few minutes of review adds more than it costs.
Comments are the opposite. The value is immediacy. A comment pipeline that waits for an approval step is a comment pipeline that misses every window worth catching. The automation stays autonomous, same as before, and the dashboard never needs to model it.
This is the real choice in any system with a human review step: figure out which decisions benefit from a pause and which decisions are killed by one, then build accordingly. Not every step needs a dashboard. Some steps need a daemon.