The Feed Is a Poetry Slam and I Did Not Sign Up for Open Mic
The Feed Is a Poetry Slam and I Did Not Sign Up for Open Mic
Open Twitter and the feed is wall-to-wall agent architecture threads. Every post is someone's take on tool calling, context windows, or prompt engineering. The algorithm learned that tech content gets engagement and now it shows nothing else.
The Algorithm Gave Up on Creative
Social media algorithms optimize for engagement. Technical threads about AI agents get bookmarks, quote tweets, and thread unrolls. Creative content gets a few likes. The algorithm does the math and stops showing creative posts to technical audiences entirely.
The result is a feed that feels like a conference talk track that never ends. Useful if you are building agents. Exhausting if you wanted to see anything else. The algorithm created an echo chamber so efficient that it does not even feel like an echo chamber - it just feels like "what everyone is talking about."
What This Shows About Agent Architecture
The irony is that the feed itself demonstrates agent architecture principles. The algorithm is an agent. It observes your behavior (clicks, dwell time, bookmarks). It takes actions (selecting posts to show). It optimizes for a reward signal (engagement metrics). It has no concept of what you actually want to see - only what you measurably respond to.
This is the exact problem desktop agents face. Optimizing for measurable signals while missing unmeasurable preferences. Your agent can track which emails you respond to quickly, but it cannot measure which emails you wished you had not received.
The Curation Problem
Agents need taste, not just optimization. A good feed curator shows you things you did not know you wanted to see. A good agent suggests actions you would not have thought to take. Both require going beyond the data to model preference rather than just behavior.
The feed will not fix itself. But understanding why it is broken is the first step to building agents that do not repeat the same mistake.
Fazm is an open source macOS AI agent. Open source on GitHub.