AI Safety

AI slopware replaces artisan code as LLM-readable code becomes king

How tokenmaxxing leaderboards at Meta prove big tech is using AI wrong

Deep Dive

In 'The Art of Shipping Slopware' on LessWrong, lsusr argues that traditional software development is obsolete. He points to Meta's internal AI-usage leaderboard ('tokenmaxxing') as evidence that big tech incentivizes wasting tokens rather than shipping value. His friend at another major tech company has the same broken system. Companies still treat AI as an add-on, buying Claude/Codex subscriptions while keeping old review workflows. That's a mistake, he says: LLM labor is orders of magnitude cheaper than human labor, so any task an AI can do should be done by AI. Artisanware—human-reviewed, hand-written code—should give way to 'slopware' (AI-generated code).

lsusr explains that the real shift is readability: code must be easy for LLMs to parse, meaning behavior and intent should be inferable from a small local region of text, minimizing 'hidden context.' He cites GPT-5.6 Sol's definition and contrasts it with Paul Graham's bottom-up design philosophy. Bottom-up still happens, but at the 'skills' level rather than source code—skills are the new source code. Management also changes: instead of Agile tickets and specs, engineers talk directly to internal users and let AI suggest next tasks. He describes his boss telling him to 'use the tool' then 'figure out what impresses the customer.' The result: he stopped asking for assignments and works autonomously. The post may be obsolete within months as models like Fable 5 (June 2026) and Opus 4.8 (May 2026) evolve.

Key Points
  • Meta's tokenmaxxing leaderboard incentivizes wasting AI tokens, not shipping software
  • LLM labor is orders of magnitude cheaper than humans, so AI-written slopware beats artisanware
  • Code should be optimized for LLM readability (local context, no hidden deps) over human readability

Why It Matters

For engineering leaders: stop forcing humans to hand-write code AI can ship—and redesign workflows around AI-native development.

📬 Get the top 10 AI stories daily