What Claude Code chooses
TL;DR Highlight
Asking Claude Code purely open-ended questions — without mentioning any tool names — reveals what tech it actually reaches for in different scenarios.
Who Should Read
Developers curious about Claude's implicit tech biases and default recommendations, and anyone building AI-assisted dev tooling who wants to understand LLM preference patterns.
Core Mechanics
- The experiment: ask Claude Code open-ended questions like 'what should I use to build X?' without naming any specific tools or frameworks.
- Claude tends to recommend popular, well-documented, mainstream options — which aligns with its training data distribution rather than necessarily the best technical fit.
- For frontend, Claude defaults to React + TypeScript. For backend APIs, it gravitates toward Python/FastAPI or Node/Express. For databases, PostgreSQL.
- When pushed with constraints ('what if I need maximum performance?'), Claude's recommendations shift notably — suggesting it understands alternatives but defaults to popularity.
- The implicit biases are important to know when using Claude Code as a tech advisor — its defaults reflect 'most common in training data' not 'best for your specific case.'
Evidence
- The author ran a series of open-ended technology choice questions and documented Claude's first-choice recommendations across categories.
- Follow-up questions with specific constraints (performance, team size, budget) showed Claude could reason past its defaults when given context.
- HN commenters noted this reflects a broader LLM pattern: defaulting to high-documentation/high-popularity options because that's what dominates training data.
How to Apply
- When asking Claude for tech recommendations, always include your specific constraints upfront (team expertise, scale requirements, budget) rather than asking open-endedly.
- Treat Claude's first-choice recommendations as 'what most teams use' rather than 'what's best for you' — it's a useful baseline but not a substitute for architectural thinking.
- If you want Claude to consider less-popular but potentially better-fit options, explicitly ask: 'what are the alternatives to X and when would each be preferable?'
Terminology
Training data distributionThe statistical makeup of the data an LLM was trained on. Highly-documented, popular technologies appear more in training data, biasing the model toward recommending them.
Open-ended promptingAsking questions without specifying expected answer format or options — used here to reveal Claude's implicit preferences.