Skip to content

Instantly share code, notes, and snippets.

Show Gist options
  • Select an option

  • Save yoyosan/839bb66a6b3259b3d606e68d044c55ab to your computer and use it in GitHub Desktop.

Select an option

Save yoyosan/839bb66a6b3259b3d606e68d044c55ab to your computer and use it in GitHub Desktop.
Summary: Thoughts on Slowing the Fuck Down by Mario Zechner

Thoughts on Slowing the Fuck Down — Summary

Source: https://mariozechner.at/posts/2026-03-25-thoughts-on-slowing-the-fuck-down/ Author: Mario Zechner | Published: March 25, 2026


Core Thesis

Coding agents are fast, but unconstrained autonomy over architecture produces unmaintainable systems faster than you can fix them.


The Three Failure Modes

1. Compounding errors without learning Humans feel pain from bad decisions and adapt. Agents don't. They repeat the same architectural mistakes indefinitely. What takes years to rot in human-written codebases rots in weeks under agent-driven development.

2. Merchants of complexity Agents make purely local decisions — no global system awareness. The result: massive duplication, mismatched abstractions pulled from training data, and a codebase that no one (including the agent) can navigate coherently.

3. Search degrades as complexity grows As the codebase bloats, agents lose the ability to find existing code. They duplicate. Which makes search worse. Which causes more duplication. A spiral.


Recommended Approach

  • Write architecture and APIs by hand
  • Use agents only for scoped, evaluable tasks (things you can verify)
  • Keep continuous code review as a discipline
  • Maintain human ownership of system design

The friction is the point — it forces understanding.


Broader Subtext

AWS outages, Windows quality regression, and general software reliability decline are symptoms of speed-over-comprehension culture, now turbocharged by agents. Each agent decision is locally rational (generates working code), but globally catastrophic (architecture degrades) — a greedy algorithm problem applied to software systems.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment