I Wasn't Automating Reading. I Was Automating the Fantasy of Catching Up
I built an AI workflow to catch up on newsletters. What it exposed instead was backlog anxiety, false obligation, and the fantasy that better summaries could solve finitude.
The pressure
I wasn’t drowning in information. I was drowning in the obligation to not miss it.
About a year ago, something felt off.
The newsletters I wanted to read had outgrown my attention. Some were paid. Many were relevant. All looked like signal.
The lived experience wasn’t abundance. It was pressure.
Every unread issue felt less like a resource and more like a claim on future time.
I wasn’t “behind” in a productivity sense. I was behind in a human one: important things arrived faster than I could metabolize them.
Unsubscribing felt like loss. Keeping them felt like debt.
The automation trap
So I tried to fix it.
Ingest everything. Classify. Summarize. Rank. Deliver a clean weekly digest.
On paper: intelligence applied to overload.
In practice: brittle.
The stack broke in different ways depending on models, prompts, and inputs. Fixing one failure mode created another. Costs shifted. Reliability never stabilized.
At one point, a summary confidently highlighted the wrong conclusion from a piece I actually cared about. Not wildly wrong. Just enough to shift the takeaway. I only noticed because I already knew the topic.
That’s the dangerous part.
That’s the surface story.
It’s not the real one.
What this was really about
I wasn’t automating reading. I was automating the fantasy of catching up.
Not comprehension first. Relief first.
Relief from backlog anxiety. From wasted subscriptions. From the sense that the important things were always just out of reach.
I thought I had a reading problem. I had a finitude problem.
I tried to solve overload at the wrong stage
I treated summarization as cleanup.
But the mistake happened upstream.
Too much entered. Too little was filtered. Noise and value mixed before any system touched it.
I didn’t need better digestion.
I needed a better gate.
The real problem is taste
A useful digest doesn’t just compress. It decides.
What deserves interruption? What can wait? What only looks important?
That’s not summarization. That’s judgment.
The system wasn’t failing to read. It was failing to know what I’d consider worth reading.
If this were just summarization, better models would fix it.
Taste doesn’t scale that way.
Where summaries break
Fluent summaries still fail where it matters: emphasis.
They drop caveats, bend intent, and keep surface meaning while shifting the point.
They don’t need to hallucinate to be dangerous. They just need to be slightly wrong.
So the work didn’t disappear. It moved from reading to verification. Without trust, the summary becomes a second inbox.
“Human in the loop” isn’t a fix
Once the machine speaks first, it shapes judgment.
First draft wins more often than we admit.
Ranking sets plausibility. Summaries set salience. Ordering sets attention.
You get slower without being safer.
The real technical failure: coordination
I didn’t hit an intelligence ceiling. I hit a coordination ceiling.
Each new layer increased failure modes. Each model change reshaped breakage. Each cost optimization added instability.
The real cost wasn’t inference.
It was babysitting.
Privacy isn’t optional
The more useful the system became, the more personal it became.
A system that reads like me also reveals me.
Which made privacy non-negotiable.
Privacy isn’t a late decision. It decides if the system can exist.
Aim lower: triage, not replacement
Don’t build a system that reads for you.
Build one that helps you choose.
Reduce junk. Surface candidates. Shape the queue. Flag what might matter.
Stop before pretending to replace judgment.
I didn’t need a system that understands for me. I needed one that knows when to interrupt me.
What changed
I aimed at the wrong level.
I tried to automate relief.
Relief is not resolution.
The unread pile isn’t a queue to clear.
It’s a reminder: more value exists than I can process.
Some things should be canceled. Some will be missed.
Maturity is cleaner choosing, not perfect processing.
The move
You don’t need a smarter system first.
You need a more honest one.
Run this once:
The Relief Audit
Would I choose this today if I weren’t already subscribed?
Does this improve my thinking, or preserve the fantasy I’ll catch up?
Does this deserve interruption, or only optional attention?
If I removed it, what would I actually lose?
Do this now.
Unsubscribe from one source today
Downgrade one to “read if time”
Keep one that truly deserves interruption
Don’t optimize everything.
Remove one false obligation.
Make one clean choice.
Clarity doesn’t start when you process more.
It starts when you stop pretending everything must be carried.