For weeks my content pipeline did exactly what I built it to do.
It pulled from a pile of sources, drafted clean articles, rendered the covers, handled the publish. The machinery worked.
And the output was wrong.
Not broken-wrong. Polished, on-topic, technically fine, and still the wrong thing to have written that day. I kept shipping articles nobody needed, on the day nobody was asking, while the thing actually worth saying sat unwritten in a folder.
It took me an embarrassing while to see it. I had automated the easy half and left the hard half to luck.
Generation was never my bottleneck.
Writing is the part everyone races to automate. Feed in a topic, get out a draft. That problem is, for practical purposes, solved. A pipeline can produce a competent article on almost anything you point it at.
So the volume went up and the value did not. More posts, same flat result. I had mistaken throughput for progress, which is the oldest trap in automation.
The draft was never the scarce resource. The decision in front of the draft was.
Selection is the job nobody automates.
Selection is choosing what deserves to exist before a single word is written. Of all the things you could say today, which one is worth your name on it.
That choice carries everything the draft cannot. Is this relevant to the people I want to reach. Is the timing right, or is the moment already gone. Does it sound like me, or like anyone. Is it the thing I am uniquely placed to say, or a thing a thousand others could say better.
A generator answers none of that. It will write whatever you hand it with equal confidence, including the wrong thing, beautifully.
Why a machine cannot pick for you.
Selection runs on context the machine does not have. What you learned last week from a client. The argument that changed your mind. The thing your audience is quietly struggling with that nobody has named yet.
That is taste, and taste is lived. It is the accumulated judgment of a specific person who has paid attention to a specific world. You can feed a model your past, but you cannot feed it your nose for what matters next.
So when I let the pipeline both choose and write, I was handing away the one part that was actually mine. The part that was the whole point.
The opinion I will defend.
Here it is. Generation is a commodity now, and selection is the entire game.
The people whose content lands are not better writers than the people whose content disappears. They are better choosers. They have a sharper sense of what is worth saying, to whom, right now. The draft is downstream of that, and the draft was never where the edge lived.
If your content is not connecting, do not reach for a better generator. Look at what you are choosing to make, and who decided it.
What changed when I treated selection as the work.
I did not throw the pipeline away. I moved myself to the front of it.
The machine still does the heavy lifting it is good at. But the choice of what to write now sits with me, on purpose, before anything runs. I look at what I actually learned this week, what my readers are actually stuck on, what only I can say with a straight face. Then the pipeline works on a target I chose, instead of a target it guessed.
The volume dropped. The hit rate climbed. Fewer articles, more of them worth reading. That trade was the whole fix, and it cost me nothing but the honesty to admit that the easy half had never been the problem.
If you are automating your content and the results feel hollow, check which half you handed to the machine. If you gave it the choosing, take that back. Let it write. You decide what is worth writing.
Your turn
What do you let a tool choose for you that you should be choosing yourself.
If this was useful
I work through this in public, the wins and the freezes both, mostly on LinkedIn and YouTube. If the real version of building in the open is useful to you, that is where it lives. LinkedIn, YouTube and X under Mirza Iqbal, and the work at next8n.com.
Top comments (1)
This hits on exactly what I've been seeing with coding agents — they execute beautifully but have no concept of 'should I be thinking or doing right now.' The selection problem you describe maps directly to the planning-vs-execution gap in AI agents. Built a small tool (Brainstorm-Mode by mehmetcanfarsak on GitHub) that enforces this separation via hooks — it puts the agent in ideation mode where it can't call tools, forcing the 'choosing' phase before the 'writing' phase. Three modes for different thinking styles. The volume drops, but the output quality goes up, exactly like you found.