Claude Cookbook

Original Article ↗ Hacker News Discussion ↗

Anthropic’s new “Claude Cookbook” of examples and skills is met with mixed reactions: some see it as a useful way to discover lesser-known features and workflows, while many argue that most “how to use AI” guides quickly become obsolete as models and harnesses absorb these patterns. Several commenters contend that elaborate prompt engineering and tooling are overhyped compared to simply asking the model clear questions, though others note that good written instructions, CLAUDE.md-style project context, and careful process design can still save tokens and reduce errors. There is also criticism of Anthropic’s frontend aesthetics examples as generic “vibe-coded” designs, mild disappointment from those expecting an actual recipe cookbook, and concern from a few about the ethics of closed AI models.

AI for Cooking and Recipes

  • Many expected “Claude Cookbook” to be a food recipe product, not a developer guide. Some express disappointment.
  • Multiple users describe using LLMs successfully for cooking:
    • Generating recipes from pantry contents or dietary needs (e.g., keto).
    • Extracting clean recipes from cluttered web pages or even TikTok videos.
    • Adjusting recipes (e.g., substitutions, sugar reduction) and tracking calories.
  • Failures do occur (e.g., instructions that omit fully cooking soaked beans), raising safety concerns about undercooked food.

Value of AI Cookbooks and How-To Guides

  • Some see “how to use AI” resources as largely unnecessary:
    • Prefer “just ask the model” rather than reading cookbooks or setting up complex harnesses.
    • Expect most clever workflow patterns to be absorbed into future models quickly.
  • Others argue cookbooks are useful:
    • They reveal features/workflows users don’t know exist.
    • Helpful to onboard less-experienced coworkers.

Prompt Engineering and Interaction Style

  • Ongoing debate:
    • One camp: modern models are “smart enough”; elaborate prompts add little over plain-language requests.
    • Another camp: clarity, precision, and good writing still significantly improve results and reduce token waste.
  • “Prompt engineering” is often reframed as basic communication skill rather than a deep technical discipline.
  • Some report that emotional or “frustrated” prompts can nudge models toward more reasonable behavior.

Claude.md, Skills, and Agent Harnesses

  • Split views on CLAUDE.md:
    • Minimalist approach: if the code is clear, extra docs are unnecessary and may waste tokens.
    • Structured approach: summary/constraints/goals in markdown reduce repeated scanning and mistakes; also used for non-coding agent tasks.
  • Many see high-churn tools (skills, MCPs, agent frameworks) as short-lived; prefer simple, durable patterns and vanilla agents.

Frontend Design and Coding Agents

  • Claude’s “frontend aesthetics” examples are widely criticized:
    • Many prefer the “before” versions; “after” is seen as cliché, gradient-heavy, and visually worse.
  • Users note agents perform better on backend or isolated components than on complex frontend state and UX, due to verification difficulties.

Ethical and Model-Openness Concerns

  • Some reject closed-model providers on ethical grounds (war use, monopolization).
  • Others note that both open and closed models can be misused, making responsibility diffuse and unresolved.