We should be more ambitious with the software we build
AI coding tools are enabling individual developers to build far more ambitious software projects, from compilers and databases to games and UI frameworks, often in weeks instead of months or years. Commenters are split between excitement about this “jet pack” effect and concern over job security, code quality, benchmark cheating, and an explosion of hard-to-maintain “vibeslopped” projects. Many argue the real bottlenecks are now product vision, process, and economics—what to build, how to differentiate when anything can be cloned, and how society will handle the broader labor impact.
Ambition, AI, and Solo “Superpowers”
- Many see LLMs as a “jetpack” that lets individuals build things they’d never have time or breadth for before (games, compilers, JS engines, note tools, infrastructure, etc.).
- Several commenters describe month‑scale, AI‑assisted solo projects that previously felt unrealistic (database tools, shops, interpreters, visualizers, retro OSes, custom hardware tools).
- Others push back that speeding up “faster ripgrep”–style work is not actually ambitious; real ambition is solving novel or human‑meaningful problems.
Jobs, Anxiety, and Adaptation
- Strong undercurrent of fear: people report layoffs, difficulty finding work, and feeling like their experience is devalued when a subscription model does “80 hours of work” in an evening.
- Debate over whether “just adapt” is realistic or cruel, especially for older workers or those living paycheck to paycheck.
- Some argue this is the same automation story other professions faced; others think AI’s breadth makes the shock larger.
- Proposals like UBI/negative income tax appear; others dismiss them as unworkable.
“Vibecoding”, Slop, and Code Quality
- Sharp split between:
- People using agents with strong review, tests, and performance profiling, claiming high‑quality results and large productivity gains.
- Critics who see a wave of fragile, overfit, hard‑to‑maintain code and expanding technical debt, especially in OSS and side projects.
- Concern that speed encourages feature bloat instead of restraint and elegance.
- Some teams report AI‑written PRs passing mostly‑automated AI reviews and deploying extremely often; others say complex codebases hit churn/complexity limits and need heavy human guidance.
Benchmarks, Trust, and Transparency (pgrust Debate)
- Lengthy dispute over a DB project accused of “cheating” a popular benchmark (hand‑rolled kernels, narrow SQL support, alleged caching).
- Project maintainer insists benchmark binaries did not use special‑case code; offers binaries and scripts for independent verification.
- Many remain unconvinced, citing: squashed mega‑commits with hundreds of thousands of files, lack of full commit history, and the need to add tests explicitly to detect AI “cheating.”
- Broader takeaway: in an era of AI‑generated systems, people demand more transparency, independent benchmarking, and formal or semi‑formal verification.
Process, Tooling, and What Becomes the Bottleneck
- Several observe that code generation is no longer the main bottleneck; coordination, product decisions, and code review are.
- Suggestions include: async or AI‑augmented review, N‑version programming with model diversity, stronger tests/specs, and better compilers/tooling to keep feedback loops fast.
- Some expect backlogs to vanish and new work to fill the vacuum; others think management will simply cut staff.