I built a machine to write for me.
It worked. That was the problem.
Over a few weeks it shipped post after post, on a schedule, in my voice, on topics in my lane. The generator never missed. And almost nobody read a single one of them. My posts average under one reaction each. The machine did exactly what I built it to do. I had pointed it at the wrong job.
If you have ever automated your content, you have met this bug. It does not crash. It does not throw an error. The output looks clean. It is quietly the wrong output, and nothing in the system tells you so.
Here is the part I am not proud of.
For weeks I tried to fix it by improving the generator. Better prompts. Tighter structure. Cleaner voice. Every change made the posts a little sharper and left them exactly as invisible. I was polishing the one part that already worked.
The pipeline reads a pile of sources, builds one brief, and produces a couple of candidate angles. I had treated those candidates as the finish line. They were the starting line. The real decision was which angle, written for whom, on a topic a real person was actually struggling with that week. That decision, selection, was the part I had automated the least and trusted the most.
Sit with that for a second, because it stops being a content problem fast.
Generation is cheap now. You can produce ten decent drafts before lunch. Your team can. Everyone's team can. So the scarce thing stopped being production a while ago. The scarce thing is choosing the one that lands. When everybody can generate, selection becomes the whole game, and selection is judgement, not throughput.
My numbers say it plainly. I averaged well under one reaction a post. The posts that win in my niche pull dozens, some of them hundreds. Same effort. Same tools. The same kind of person at the keyboard. The gap was never the writing. It was choosing a thing somebody actually needed to read.
The threads that pull real argument in my world right now are plain and unpolished. A developer asking whether it matters that they stopped improving in their free time. Someone wondering if AI made their team less collaborative. No banner, no structure, full of replies. My pipeline had produced nothing that week a person would argue about. It had produced things. None of them had a pulse.
Here is the opinion I will defend. More generation makes this worse, not better.
Every extra draft your pipeline produces is one more thing competing for the single slot that matters this week, and a weak selector drowns in its own volume. It mistakes activity for progress. It measures how much it made and never asks whether any of it mattered. A pipeline that produces less but chooses well beats one that floods the queue, every time.
You can feel this outside content too. The codebase that ships features nobody asked for. The backlog that grows faster than anyone can decide what is worth doing. The design tool that hands you forty variations and no help picking one. In every case the machine got better at making and no better at choosing, and the bottleneck quietly moved while we kept optimizing the old one.
So the question I had to answer, and the one I think most automated pipelines dodge, is a hard one. How do you know which thing is worth someone's attention before you spend it.
When I started grading my pipeline on that question instead of on output, things changed. I stopped asking did it generate a clean post. I started asking did it pick a topic a real person was hurting on this week, and did it have a reason beyond the draft reading nicely. The generator barely changed. The selection got a spine. That was the part that had been missing the whole time.
I wish I could tell you I designed it that way from the start. I did not. I built the impressive half first, the half that produces things, because producing things feels like progress. The quiet half, deciding what deserves to exist at all, is the half that actually decides whether anyone shows up.
If you are building anything that can now generate faster than you can wisely choose from, that is where your effort belongs. Not a better generator. A better selector. The uncomfortable truth is that generation stopped being the constraint for most of us, and most of us are still pouring our days into it anyway.
My machine is still running. It produces less now, and it argues with itself more about what is worth publishing. That trade has been the only thing that moved the numbers at all.
Your turn
What is the last thing you built that nobody needed. Be honest.
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. Find me on X, GitHub, and the work at next8n.com.
Top comments (1)
Generating code or content with AI is basically free now. The real craft lies in choosing what actually deserves to exist out of hundreds of variations. Selection is judgment; the rest is just system noise.