Skip to content

Complexity as Enemy

mario-zechner: "My biggest enemy is complexity — it's also your agent's biggest enemy." A practical restatement of a classic engineering maxim, but with a 2026-specific twist: the agent's effective context window sets a hard ceiling on how much complexity it can reason over per task.

The context-window argument

"If I have a 600k LOC codebase and my agent can at best be effective up to a context window of around 200k tokens, how much of the code can the agency see? A third. If you manage to get all the relevant code for a task into that context window, you're probably okay."

Corollary: rejecting features that don't fit the current architecture is the single highest-leverage lever for keeping an agent productive in a real codebase. Mario explicitly: "Saying no to a new feature that's currently not possible with the current architecture is the one thing that keeps the codebase quality high and the complexity low."

This cuts directly against the industry posture of "burn as many tokens as possible" — more tokens only help if the relevant code is in the window at all.

Why agents make it worse, not better

Armin's counterpoint: the cognitive cost of accepting a feature used to be a natural governor. When you had to type it yourself, you'd push back more often. Agents remove that cost → "You say yes, I want this and that. I want this and I want this because I don't have to type it myself, I don't have to think about it. Good enough." Complexity compounds silently.

Connects to