My experience with Claude Code after two weeks of adventures

Original Article ↗ Hacker News Discussion ↗

Enthusiastic Views and Use Cases

  • Several commenters say Claude Code (CC) feels like “changing jobs” or working with a capable junior/mid engineer: great at boilerplate, CRUD, refactors, debugging, and greenfield apps in mainstream stacks (TS/Next.js, Laravel, Ruby, Python, Go).
  • Reported wins include: Airflow DAGs, billing dashboards, cloud cost analyzers, DB rebuild tools, mobile apps with nontrivial features, and data‑migration pipelines, sometimes in days instead of weeks.
  • People like using CC to overcome “white page” inertia, explore unfamiliar tech, and offload tedious work while keeping interesting problem‑solving for themselves.

Negative Experiences and Limitations

  • Others find CC slow, fragile, or unusable on complex, legacy, or niche codebases (C/C++ systems, Win32, Haskell/Bazel/Flutter, Godot, etc.).
  • Common failure modes: incomprehensible or redundant code, runaway edits, “grep hell”, invented APIs, non-durable designs, and gaslighting-like insistence on nonexistent functions.
  • Visual/UI tasks and subtle design details (e.g., a wrong-looking SVG icon) often slip past its “self‑verification”.

Best Practices and Workflow Patterns

  • Strong consensus that success depends heavily on:
    • Good tests and fast feedback loops; otherwise it may “fix” tests instead of code.
    • Structured codebases with clear conventions.
    • Per-folder CLAUDE.md / repomaps, explicit architecture and business rules.
    • Using plan mode, small tasks, explicit constraints, and having it keep a journal of issues/decisions.
  • Many treat it as an intern: you must review every change “like a hawk” and be ready to discard bad runs.

Comparison with Other Tools

  • Cursor: praised for in‑editor UX and tab completion; some find Cursor+Claude better than CC alone, others say CC’s agentic behavior and tool use are clearly superior.
  • Other tools mentioned: Cline, Aider, opencode, RepoPrompt, Kimi‑K2, Gemini 2.5, GPT‑4.1/o3, Grok 4; quality varies by model and task.

Impact on Developers and Jobs

  • Strong debate over whether CC makes juniors unnecessary and eventually lets a few seniors oversee AI‑generated code for many.
  • Pushback that CC doesn’t learn, repeatedly makes the same mistakes, and that mentorship is an investment in future seniors.
  • Some seniors report real speedups (1.5–10× by their own metrics); others cite studies or experience where perceived gains hide slower real progress.

Costs, Limits, and Hype

  • Rate limits on Pro vs Max tiers frustrate heavy users; some route through Bedrock or APIs.
  • Worry that current prices are “Uber‑era subsidies” and may spike later.
  • Many complain that rave posts lack concrete details (stack, size, tests, diffs), fueling suspicion of influencer hype and calling for more rigorous, specific case studies.