Tag
The article argues that as AI makes code generation cheaper, software engineering's value shifts from writing code to specifying, supervising, validating, and owning the resulting system, favoring engineers who adopt AI tools while raising verification standards.
A quote from Florian Herrengt about how AI-generated code can lead to convoluted projects where no one understands the system, highlighting risks in software engineering.
Linus Ekenstam discusses the gap between AI writing 75% of Google's new code and only 10% measured velocity gain, and how Sonar aims to close that gap.
A security scan of 630+ vibe-coded sites reveals common vulnerabilities such as missing CSP, insecure cookies, and exposed API keys, highlighting risks of AI-generated code.
Twitter user @corbin_braun claims Fable 5 generated 1.2 million lines of code for his stealth software, suggesting a major advance in AI-powered coding.
A large-scale empirical study analyzing 3.52 million C++ code changes in production to compare AI-generated versus human-written code quality, finding higher coupling and compute overhead but showing targeted feedback can mitigate issues.
The blog argues that as AI generates code faster, understanding the combined execution becomes difficult, and proposes using graph engineering with a compiler to create deterministic orchestrators.
v0 can now turn a full Figma file into a working app by exploring pages and frames, then building the screens as one app.
An article arguing that AI tools eliminating the difficulty of writing native iOS code have removed the last major moat in software, making mobile app development wide open after being neglected in favor of web apps.
Palmer Luckey predicts that AI-generated code will be surprisingly well-optimized, resembling John Carmack-style machine code, countering fears of inefficiency.
This paper introduces generative compilation, a method to obtain compiler feedback on partial programs during AI code generation, using a 'sealor' transformation that enables standard compilers to diagnose incomplete code. Evaluated on Rust coding tasks, it reduces non-compiling outputs and improves functional correctness by catching errors early.
Linus Torvalds is reportedly using AI to write a Python visualizer for a new project, signaling a potential shift in traditional software engineering practices.
A tweet by Jerry Liu highlights the risky practice of running hallucinated code, referencing a technique where Fable costs are cut by converting code into an image for OCR processing.
Gergely Orosz argues that experienced software engineers remain valuable because they can discern good code from bad, something AI code generators still cannot reliably do.
Fable 5 model generates a 644-line single-file HTML from a one-sentence description, achieving a realistic American flag fluttering effect including folds, lighting, and rhythm, all in one shot without any revision, demonstrating the AI's capability in complex code generation.
Fable 5 integrates with Composer 2.5 Fast for enhanced AI code generation capabilities.
The article explores how AI code generation excels at local code chunks but struggles with global program understanding, leading to excessive defensive checks. It examines whether programming language design can help, using an example of local reasoning assuring global properties.
The author argues that un-monitored AI code generation ('vibe coding') creates compounding technical debt, and proposes an 'AI-Powered Developer Manifesto' advocating for macro-level architectural control.
Charlie Marsh and Ryan Peterman discuss the impact of AI writing code on software engineering, arguing that PR review costs remain constant while generation costs are zero, and share experiences building Ruff, emphasizing rapid iteration, authentic marketing, and the importance of performance benchmarking.
Discusses the emerging pattern of using external harness loops to extend AI coding agent sessions beyond normal boundaries, and critiques current code quality issues.