You're Not a Better Engineer Because You Type Git Commands by Hand
Overall reaction to the article’s stance
- Many see the title as flamebait or prescriptive (“you’re burning your life”), which pushes people away from AI instead of persuading them.
- Several commenters agree with delegating boring, mechanical tasks to AI, but object to the absolutist life-advice framing.
- Others feel the post lacks concrete examples or evidence for claims like “AI does this job better,” relying too much on “trust my X years of experience.”
AI for Git operations and PR workflow
- Some engineers heavily delegate git work (rebases, merges, branch cleanup, PR labeling, checklist updates) to agents, calling most of it mechanical overhead.
- Others keep LLMs away from direct git execution due to trust/safety concerns, using them only for advice or one-off commands.
- A recurring view: AI-driven git is safe and effective if you already understand concepts like rebases, merges, and fast-forwards; dangerous if you don’t.
Commit messages, comments, and communication
- Strong pushback on AI-written commit messages and PR descriptions:
- Often verbose, irrelevant, or miss the actual intent.
- Degrade git history, making bisecting and code archaeology harder.
- Counterpoint: LLMs are excellent summarizers when carefully prompted, and can enforce consistent house style across a team.
- Several argue that manually writing commits forces you to understand and take responsibility for changes, which improves overall technical quality.
Outsourcing thinking vs. abstractions
- Concern that delegating too much to AI leads to atrophied skills and less understanding of one’s own code; “outsourcing thinking” vs. “tooling away drudgery.”
- Others reply that abstractions are a cognitive necessity; AI is just another abstraction layer, like higher-level languages or IDE refactors.
- Common middle ground:
- You should understand at least one layer below where you operate.
- AI is fine for grunt work if you still read diffs, run tests, and apply judgment.
Skills, productivity, and future of engineering
- Some report AI-heavy workflows don’t dramatically speed feature delivery but free time for design, security, and data modeling, with fewer bugs.
- There’s worry that if a role is reducible to “pressing next” on an AI/IDE, it’s easily replaced or commoditized.
- Low-level knowledge (git, CLI, Linux) is defended as valuable when things break, though others predict models will eventually handle even recovery tasks.
Meta-discussion: AI content and detection
- Visible frustration with constant accusations that every article is “AI slop” and overreliance on AI detectors.
- Suggested norm: judge content by usefulness and clarity, not by whether it was AI-assisted.