▸ T0 · main thread · write Substack post from build log · interactive
Running 12 LinkedIn posts a week hurt more than it helped. Two posts a day competed for the same golden hour window, and the feed ranker almost never shows both to the same person. The second post did not compound the first; it cannibalized it.
So I cut to one post a day, seven a week, keeping the late morning slot with the highest reach. Weekly output dropped roughly 40% but useful output, meaning posts that actually reached the intended audience, likely went up.
The second problem was subtler. Every post was Claude this, agentic systems that. Because all the signal sources feeding the engine were AI and dev news, 100% of the topic mix landed in one niche. Founders and students following the account saw a single note played on repeat.
The fix was a new topic type: reflection. 36 evergreen seed ideas covering lessons, tradeoffs, and observations that are not AI content, stored in topics_reflection.json. The mixer now slots reflection at roughly 2 out of every 6 posts, around 28% of total volume, while AI content stays the 2 in 3 majority. The rotation cycles oldest first so seeds do not repeat until the full pool has run.
The lesson keeps proving itself: more output is not more reach. The feed algorithm is not your amplifier; it is a filter. Post twice into the same time window and the algorithm picks one and buries the other. You did the work twice for one result. The cadence that feels productive is often not the cadence that performs.
Topic mix matters for the same reason from a different angle. A monothematic account trains its audience to expect exactly one thing, which means the audience it attracts is small and the audience it could have had never shows up. Adding a reflection stream is not losing focus; it is finding the full shape of what the account is actually about.
189 tests pass on the new config. The proof will be in next week's numbers.