We should be more ambitious with the software we build
Dan Luu argues that AI fundamentally shifts the boundaries of what's possible in software, allowing developers to pursue ambitious projects previously deemed unfeasible due to time or complexity. He highlights how AI empowers individuals to build sophisticated tools and systems with unprecedented speed, despite the common perception of 'AI slop.' The Hacker News community debates this, wrestling with the promise of enhanced productivity versus concerns about code quality, job displacement, and the true meaning of ambition in the age of AI.
The Lowdown
Dan Luu, building on Michael Malis's assertion, posits that developers should be more ambitious with the software they create, as AI tools have dramatically increased individual capabilities. He illustrates this with personal projects, such as a custom regex engine, a faster ripgrep fork, and a superhuman AI for the board game Azul, all built in significantly less time than previously imaginable.
Key points from the article include:
- AI enables the rapid development of "works for me" quality software, demonstrating how previously daunting tasks (e.g., compilers, verification) can now be tackled by individuals with minimal domain expertise.
- Luu acknowledges the widespread skepticism and prevalence of "vibeslop" in AI-assisted projects but stresses that diligent testing and a systematic approach can still yield high-quality, ambitious results.
- He cites examples of AI's improved capabilities, like GPT-6 Astra's success with complex tasks like zstd implementation and IMAP RFC interpretation, which were beyond earlier models.
- An appendix features Yossi Kreinin's cautionary perspective, warning that most people should be less ambitious with AI to avoid "abject failure" due to models being black boxes and generating messy, unmaintainable code.
- Luu counters that setting up self-checking systems (akin to fuzzing or property-based testing) can make ambitious AI projects viable, even for complex domains.
- He shares further examples, including Peter Geoghegan's AI-generated Neovim code formatter for highly specific needs, and Anthropic's internal tool replacement, to underscore the breadth of what's now achievable.
The author concludes that while AI doesn't make everything "just work," it expands the horizon of what an individual or small team can achieve, encouraging a greater degree of ambition in software development.
The Gossip
Ambitious Advances & AI Assistance
Many commenters echo the article's sentiment that AI acts as a powerful accelerant, enabling them to pursue projects previously deemed too complex or time-consuming. From building intricate database tools like `walshadow` to significantly faster compilers and game remakes, individuals share how AI has become their 'jet pack,' dramatically boosting productivity and allowing for the creation of innovative, personalized software. There's a clear sense that the new era of AI empowers engineers to tackle problems on a grander scale.
Quality Conundrums & "Slop" Skepticism
A significant thread discusses the challenges of AI-generated code, often characterized as 'vibeslop.' Commenters express concern that AI can produce messy, overfit, and unmaintainable code, especially when users lack deep understanding of the problem or fail to review the output meticulously. The debate touches on the practicalities of moving beyond an MVP, the need for human expertise in navigating complex tradeoffs, and direct accusations of certain projects, like `pgrust`, engaging in benchmark 'cheating' due to AI's reward-trained nature. This perspective highlights the tension between speed and rigorous quality control.
Future of Labor & Existential Angst
The discussion delves into the broader societal and personal implications of AI for software developers. Some acknowledge the 'misery' AI causes by threatening job security, especially for older or less 'creative' programmers, and criticize what they see as tribalism from those who dismiss these concerns as mere 'complaining.' Others argue that this disruption is simply part of a larger historical pattern of technology impacting various professions and that adaptation is necessary. The conversation also touches on ethical dilemmas, such as the cost of AI tools and the moral implications of feeding AI labs with open-source contributions.
Verification Ventures & Technical Nuances
Commenters dive into specific technical challenges and solutions related to ambitious AI-assisted projects, particularly in formal verification. The creator of `pgrust` details his work using a custom compiler to translate C/Rust into Lean for formal proofs, struggling with context scaling and exploring orchestrators for managing complex proofs. This leads to questions about the reliability of such translation compilers and comparisons to other verification approaches, like reasoning at the RISC-V level, showcasing the deep technical interest in pushing AI's boundaries in software assurance.