ai-assisted programming

Bilingual

If AI Writes All the Code, What Do the Programmers Do?
2 months ago
- The author's code writing has shifted from 90% human-written to 90% AI-written in eight months.
- The task involved optimizing a matrix-multiply kernel using Cluster Launch Control (CLC) on GPUs, where AI assisted in finding, understanding, fixing, and benchmarking code.
- AI helped identify unused shared memory to fix a deadlock, and suggested multiple optimizations, leading to faster code for problematic inputs.
- The overall speedup was modest (4 days vs 5 days) because non-coding tasks like benchmarking, reviewing, and understanding remained time-consuming.
- AI enabled broader exploration of optimizations and produced higher-quality results with better plots and bug detection via automated reviews.
- A downside is reduced understanding of the codebase compared to doing it manually, but the author still needed to make final decisions and simplify code.
- The author finds satisfaction in shipping more features quickly, but feels using AI is like a cheat code in a video game—hard to justify not using it for efficiency.
- AI still requires human oversight for long-term maintainability, judgment calls on trade-offs, and to prevent unhinged behavior in pursuing goals.
<antirez>
2 months ago
- The author feels a responsibility to help less experienced programmers adapt to AI-driven changes in programming, based on their own foresight and experience.
- The key shift is that programmers should focus on controlling the ideas and design of software rather than reading code line-by-line, because LLMs handle local optimization well and generate large amounts of code.
- Reasons to prioritize ideas: LLMs produce verbose code, humans have limited time, and it's more efficient to evaluate design through prompts instead of scanning functions.
- The author's work on DwarfStar and comparison with other implementations shows that AI can help catch errors in complex domains like local inference, where hand-written code is often flawed.
- Resistance to AI may be ideological; rigorous engineering and testing are more valuable than manual code review, especially for domains like GPU kernels.
- Even for codebases like Redis, detailed code review is becoming less useful; instead, documenting design ideas in a DESIGN.md file would be more valuable for future modification.
- Young programmers should learn fundamentals (e.g., writing a small interpreter) but not waste time reviewing LLM-generated code for mundane tasks.
- The software world was already 'rotten' before AI, and AI offers opportunities to improve it, though the change is painful.
The Tower Keeps Rising
2 months ago
- The Tower of Babel story illustrates that technological progress relies on coordination through a shared language, not just tools or knowledge.
- Large software projects are limited by coordination and shared understanding, not just individual coding speed.
- The shared language in software includes common understanding of concepts, boundaries, invariants, ownership, and system design, often maintained through documentation, code reviews, and conversations.
- AI-assisted programming removes friction in code changes but risks eroding the shared architectural understanding among developers.
- In AI-assisted engineering, construction can continue even as shared understanding collapses, unlike in the Babel story where construction stops, leading to a disorienting lack of immediate failure.
Software Bonkers
2 months ago
- The author has been deeply engaged in building software, driven by strong opinions about software being fast, knowable, and owned.
- Using Claude Code, the author rebuilt Twitter with features like disappearing posts, limited posts/replies, and non-algorithmic timelines, fostering a positive community.
- The author created a custom membership platform atop Memberful, adding tools like auto-generated chapters for livestreams and a searchable video archive.
- Small personal utilities were built, such as tools for appending text to files and live currency conversions, used daily.
- In a tax season project, the author built a custom accounting software named TaxBot2000 in five days to handle complex multi-currency, multi-country finances.
- TaxBot2000 is fast, local, handles CSV imports, categorizes expenses, reconciles wire transfers, and adapts through learning, replacing tools like Quicken.
- The software reflects the author's unique financial situation and scarcity mindset, providing a holistic dashboard for transparency.
- The author envisions a future where AI-assisted programming becomes more fluid, enabling personalized software and reducing reliance on proprietary SaaS.
- There's an emphasis on control over data and gratitude to the open-source community, highlighting an epochal shift in software creation.

ai-assisted programming

If AI Writes All the Code, What Do the Programmers Do?

<antirez>

The Tower Keeps Rising

Software Bonkers