Simplicity in the Age of AI-Assisted

2026-05-14 · page · source summary

The argument that LLMs don't change what good programming looks like, but they change our relationship to simplicity. The real unlock isn't faster generation -- it's the ability to cheaply rebuild systems without inherited complexity. When humans identify unnecessary complexity, LLMs make acting on that judgment nearly free.

Key Quotes

"The hard part is the human part: recognizing which complexity is essential and which is inherited."

"The layers exist for communication, not computation."

"The complexity justifies the roles. The roles produce the complexity."

"The LLM will not question your architecture."

"What LLMs actually give you is the possibility of simplicity: the removal of complexity that shouldn't have been there."

Key Themes

#simplicity #cognitive-debt #agentic-coding

The argument unfolds in four moves:

  1. Complexity through linguistic layers -- a single feature (sticky header toggle) gets expressed across user stories, CSS, JS, config files. Each layer serves human communication, not computation.

  2. Complexity explosion -- add persistence and the feature balloons into databases, migrations, APIs, Kubernetes, monitoring, specialized roles. "All of this to remember whether the header should be sticky."

  3. Self-perpetuating cycles -- infrastructure creates roles that justify the infrastructure. "The complexity justifies the roles. The roles produce the complexity."

  4. LLMs inherit, not question -- models reproduce existing patterns because they were trained on millions of repos that perpetuate those patterns. They won't tell you your architecture is unnecessarily complex.

The punchline: the value of LLMs isn't that they generate code faster. It's that they make it cheap to throw code away and rebuild simpler. Once a human makes the call that complexity is inherited rather than essential, the LLM can regenerate a simpler version in minutes instead of weeks of careful refactoring.

This is the philosophical foundation for several other pieces: Write Only Code (if code is disposable, simplicity is the only durable value), The Claude C Compiler (Lattner's point about investing in structure), Prefix Effects (if early patterns solidify, start simple), and Elements of Code (comprehensibility as the primary virtue). Grug Brain Developer got there earlier and funnier: "complexity is spirit demon that enter codebase through well-meaning but ultimately very clubbable non grug-brain developers." Same diagnosis, different vocabulary — grug's "spirit demon" is this article's "inherited complexity."

Critical Analysis

This is one of the sharpest pieces in the batch. The four-move argument is tight, the quotes are memorable, and the conclusion is both counterintuitive and correct: LLMs are more valuable as demolition tools than construction tools.

The limitation: the author doesn't address when inherited complexity is load-bearing. Not all organizational infrastructure is waste -- sometimes the Kubernetes cluster exists because you actually need multi-region deployment, not because a DevOps team justified its existence. The hard judgment is distinguishing load-bearing from vestigial, and the article acknowledges this is the hard part without offering a method for making the call.

Sources: Simplicity in the Age of AI-Assisted Last updated: 2026-05-14