How to reduce AI slop in your codebase effectively

TL;DR

Attention Span fixes how the AI talks and reduces noise in explanations. Karpathy Skills constrain AI judgment to avoid overengineering. Anti Slop provides lint-like rules to prevent sloppy code, while Verification Before Completion mandates evidence before claims. Together these tools cut code bloat and improve maintainability in AI-assisted coding.

Transcript

The guy that invented the phrase by coding doesn't like the code it produces. Andre Karpathy says agents bloat abstractions, have poor code aesthetics, copypaste code blocks everywhere and make a mess. He also says models make wrong assumptions on your behalf and just run with them. They leave dead code everywhere and they write a,000 lines when a ... Read More

Key Insights

• Attention Span reduces AI verbosity by enforcing predefined output styles such as status checklists and concise summaries.
• Karpathy Skills advocate thinking before coding, simplicity first, matching the existing style, and goal-driven execution to cut unnecessary AI decisions.
• Anti Slop introduces a llinter-based rule set to flag AI induced issues like vague parameters and unnecessary conditional object spreads.
• Verification Before Completion enforces evidence and proof commands before claims of completion, cutting false positives from AI outputs.
• The four tools together reduce the cognitive load of reviewing AI code by taking away judgment calls from the AI and placing them in the language and tooling.
• Anti Slop can be installed as a plugin or linked as rules, making the AI adhere to team-specific coding standards.
• Attention Span files are lightweight markdown guides that shape how an AI should respond, improving readability and planning.
• The Kotlin angle shows how reducing AI complexity can also cut lines of code, with Kotlin a focal point for cleaner, testable code.

Questions & Answers

Q: What is Attention Span in AI coding

Attention Span is a collection of markdown files that tell the AI how to answer, installed as output styles in cloud code or linked in agents.mD, designed to reduce verbose and unreadable outputs. It makes planning and status updates concise, enabling faster reading and easier comprehension of AI responses.

Q: What are Karpathy Skills

Karpathy Skills are four rules to manage AI behavior during coding: think before coding, do not assume when a request is ambiguous, simplicity first by changing only the necessary lines, and goal-driven execution by giving the AI explicit success criteria. These rules limit unnecessary complexity and prevent overengineering.

Q: What does Anti Slop aim to fix

Anti Slop provides a set of rules that flag AI induced coding issues, like vague parameters and problematic conditional object spreads. It makes the agent run lint checks to ensure code clarity, enabling deterministic feedback and enabling the team to enforce consistent code quality across AI generated outputs.

Q: What is Verification Before Completion

Verification Before Completion is a skill that requires evidence before the AI can claim done, fixed, or passing. It mandates proof steps and commands to validate changes before releasing, reducing the risk of untested or incorrect AI generated code slipping into production.

Q: Why are rules like anti slop important

Rules like Anti Slop are important because they remove subjective judgment from the AI by enforcing objective criteria, such as specific parameter definitions and explicit types. This approach minimizes ambiguous or bloated code and helps reviewers understand what the code expects and how it should behave.

Q: How does Attention Span affect planning sessions

Attention Span improves planning sessions by presenting implementation plans in a skimmable format, enabling faster review and catching potential slop before coding begins. This enables teams to align on scope and expectations, reducing fatigue and improving confidence in the code plan.

Q: How can a team install Anti Slop

A team can install Anti Slop by adding the rules manually or using a dedicated agent skill to apply them. The rules are designed to reflect a developer's taste and can be customized to fit the team’s standards, providing a consistent baseline for AI generated code.

Q: What role does the compiler play in code verification

The compiler plays a role by guaranteeing that code adheres to language rules and behaves consistently, removing many judgment calls from AI generated code. While the AI can still introduce issues, the compiler enforces fundamental correctness and consistency across sessions and codebases.

Summary & Key Takeaways

• Attention Span reduces long, run-on AI outputs by using structured styles, making planning and status updates skimmable and clearer within seconds.

• Karpathy Skills provide four rules that force simplicity, surface ambiguity, and goal-driven execution to minimize unnecessary complexity in code produced by AI.

• Verification Before Completion requires concrete evidence before declaring tasks done, ensuring changes are actually validated before release and increasing trust in AI-generated code.