{ "slug": "token-maxing-is-what-you-do-when-you-cant-budget", "title": "Token maxing is what you do when you can't budget", "date": "2026-05-12", "featuredClaim": "Token maxing is what you do when you can't budget—name your constraint, or every AI choice is just a vibe.", "description": "The HTML vs Markdown debate for AI-generated artifacts is really a symptom of a deeper problem: teams defaulting to maximum token usage without naming their constraints. The article argues for token budgeting—naming a primary constraint and letting format and model choices follow—backed by evidence from ETH Zurich, Anthropic, and Stanford/MIT studies.", "keyPoints": [ "The HTML vs Markdown fight was two teams optimizing for different constraints (cognitive density vs token cost), not a methodology battle", "An ETH Zurich study of 438 coding tasks found LLM-generated AGENTS.md context files cut success rates by 3% while raising costs over 20%", "Context rot degrades model performance at every length increment, and models can't predict their own token spend (up to 30x variance)", "Budgeting means naming a primary constraint in one sentence and measuring costs—maxing without naming the trade is just defaulting" ], "topics": [ { "id": "strategy", "slug": "ai-strategy", "label": "AI Strategy", "description": "Strategic planning and implementation approaches for AI adoption" }, { "id": "measurement", "slug": "measuring-ai-roi", "label": "ROI & Measurement", "description": "Measuring AI impact and return on investment" }, { "id": "implementation", "slug": "ai-implementation", "label": "Implementation", "description": "Hands-on implementation techniques and frameworks" } ], "claims": [ "An Anthropic engineer argued HTML should replace Markdown as the default format for AI-generated artifacts.", "Mintlify reported a thirty times reduction in token usage when serving documentation as clean Markdown.", "An ETH Zurich study found LLM-generated context files cut task success rates by three percent.", "Chroma tested eighteen frontier models and found performance degrades at every context length increment.", "Token maxing without naming a primary constraint is defaulting dressed up as strategy, not deliberate choice." ], "claimTitles": [ "Anthropic Engineer Backs HTML", "Mintlify Chooses Markdown", "Context Files Backfire", "Context Rot Is Universal", "Maxing Without Constraint Is Defaulting" ], "originalUrl": "https://aiadopters.club/p/token-maxing-is-what-you-do-when", "claimProvenance": [ "source-summary", "source-summary", "source-summary", "source-summary", "author-interpretation" ], "primarySources": [ { "title": "an engineer on the Claude Code team at Anthropic", "url": "https://thariqs.github.io/html-effectiveness/", "publisher": "thariqs.github.io", "claimIndices": [ 1 ] }, { "title": "Mintlify shipped the opposite default", "url": "https://www.mintlify.com/blog/context-for-agents", "publisher": "mintlify.com", "claimIndices": [ 2 ] }, { "title": "researchers from ETH Zurich’s SRI Lab published a study", "url": "https://arxiv.org/abs/2503.13501", "publisher": "arxiv.org", "claimIndices": [ 3 ] }, { "title": "Chroma tested eighteen frontier models in July 2025", "url": "https://research.trychroma.com/context-rot", "publisher": "research.trychroma.com", "claimIndices": [ 4 ] } ], "quote": "If you can't name the trade in one sentence, you're not maxing. You're defaulting and dressing it up.", "keyStatistics": [ { "stat": "Up to 30x variance in token spend", "context": "Bai et al. at Stanford and MIT found running the same agent on the same task produces a thirty-fold range in cost, with models predicting their own usage at correlations no higher than 0.39." }, { "stat": "3% lower success, 20%+ higher cost", "context": "ETH Zurich's SRI Lab study of 438 real-world coding tasks found LLM-generated AGENTS.md files cut task success rates by three percent while raising costs by over twenty percent." }, { "stat": "80% fewer tokens served as Markdown", "context": "Cloudflare measured an eighty percent token reduction when serving the same blog page as Markdown instead of HTML, three weeks after Mintlify reported a thirty times reduction." }, { "stat": "15x token multiplier for multi-agent research", "context": "Anthropic's multi-agent research system burns fifteen times more tokens than chat, a cost the team accepted deliberately after naming breadth-first parallelism as its primary constraint." } ], "supportingContext": "The article grounds its argument in peer-reviewed and industry research rather than ideology, citing controlled studies from ETH Zurich, Stanford and MIT, and Chroma's testing of eighteen frontier models. Its methodology is to reframe the visible HTML-versus-Markdown debate as a symptom of a deeper failure: teams defaulting to maximum context, richer formats, and more agents without naming the constraint they are optimizing against. For practitioners, the actionable takeaway is a three-test diagnostic—name your primary constraint in one sentence, measure per-task costs and variance, and identify what you would give up to relax the constraint by half. Teams that pass these tests can then deliberately choose maxing where the trade is worth it, supported by caching infrastructure that cuts input costs by up to ninety percent. The core skill being advocated is asking what token spend actually buys, at what scale that calculation flips, before any format or model decision follows.", "canonicalUrl": "https://kbanc.com/claims-library/token-maxing-is-what-you-do-when-you-cant-budget", "markdownUrl": "https://kbanc.com/md/claims-library/token-maxing-is-what-you-do-when-you-cant-budget.md", "jsonUrl": "https://kbanc.com/api/claims/token-maxing-is-what-you-do-when-you-cant-budget.json", "source": { "publisher": "AI Adopters Club", "title": "Token maxing is what you do when you can't budget", "url": "https://aiadopters.club/p/token-maxing-is-what-you-do-when" } }