Building a Mac app with Claude code

LLMs as Skill Amplifiers vs Replacements

• Many see tools like Claude Code as strong amplifiers of existing expertise, not replacements.
• Experienced engineers report jumping between stacks (Python → Go, Swift, Bash, AppleScript, React, etc.) far more easily because LLMs handle “trivia” (syntax, boilerplate, API usage).
• Several argue non-technical users still struggle: without conceptual understanding, results are slow, brittle, or disastrous.

Learning, Expertise, and “Vibe Coding”

• One camp says you “can’t acquire expertise” if the LLM does the coding; you learn by solving problems yourself, not by consuming solutions.
• Others claim they’ve learned a lot from LLM examples (including overlooked library features) and that interns need far less basic help, freeing mentors to focus on deeper topics.
• There’s debate over whether reading generated code grows real understanding or just builds a shaky pattern library. Some call LLM learning “a land of mirages.”
• Concerns about “vibe coding”: unskilled devs iterating until errors disappear without grasping root causes, embedding subtle bugs they can’t reason about.

Code Quality, Context, and Limitations

• LLMs praised for small tools, focused API examples, configs, refactors, docs, and commit/PR messages.
• Multiple reports that quality drops sharply on larger, complex systems: context windows overflow, the model contradicts earlier architecture, and suggestions become incoherent.
• Users stress the need for human review, strong mental models, and tests; some feel their skills atrophy if they lean too hard on the tool.
• Hallucinations and factual errors remain a blocker for those who want zero incorrect output.

Pricing, Access, and “Enshittification”

• Claude Code’s higher tiers ($100–$200/month) spark pushback, especially compared with traditional dev tools and for hobbyists or low-income learners.
• Worries: AI firms aren’t profitable, so prices and quality may worsen once investor pressure mounts; parallels drawn to YouTube, Uber, Netflix post-growth.
• Counterpoints: token prices are generally trending down; enterprise revenue may subsidize costs; open-source models (e.g., Llama) and local deployments could mitigate lock-in—though hardware requirements create another access gap.
• Some fear a “haves vs have-nots” job market if serious productivity requires paid AI subscriptions.

Workflows and Use Cases

• Many describe a new workflow: terminal + Claude Code + a simple editor (Neovim/Emacs) instead of heavy IDEs.
• Reported uses: Mac/iOS apps, CLI tools, config rewrites, AWS infra, Mac utilities, classic Mac OS experiments, WordPress plugins.
• IDE/agent integrations typically propose diffs rather than writing directly, with git or local history as safety nets.

Emotional Impact and Future of the Profession

• Several experienced developers feel simultaneous “superpowers” and sadness: work feels more like industrialized assembly than craft.
• Others frame this as another abstraction leap, akin to moving away from assembly or hand tools; craftsmanship persists at deeper layers.
• Educators and seniors worry about how to advise students: 4-year CS degrees vs rapidly evolving tools, and whether beginners can still build durable fundamentals in an AI-heavy environment.