Tag
This article challenges the narrative that AI has solved coding by emphasizing the critical roles of maintenance, reliability, and accountability in software engineering beyond code generation.
The paper critiques AI models like vibe coding for reducing programmer understanding and harming software engineering training, highlighting ethical concerns and the need for careful AI use.
The article lists 10 open-source GitHub projects that turn AI models into coding engineers, emphasizing the iterative loop of code development and debugging.
A social media post expressing relief that the era of hand-typing code is over, likely pointing to advancements in AI coding assistants.
This paper studies developer productivity at Google, finding that code quality and other factors causally affect productivity, with code quality showing a strong impact through panel data analysis.
The paper introduces WideSWE, a benchmark for evaluating coding agents on cross-repository tasks, based on 120 real-world tasks, showing that agent success rates vary and joint execution can improve performance.
The tweet discusses using Jev as a fuzzy linter to handle complex coding guidelines that are not suitable for deterministic linting, with examples like useEffect and comment management.
DHH reveals that 37signals has largely shifted to AI-assisted coding with agents, increasing code output, and questions the ongoing debate about AI's effectiveness in real coding tasks.
Gergely Orosz contrasts engineering mindset with hacker mindset in AI coding, and Simon Willison notes that coding agents require extra discipline for software engineering.
Anthropic engineers outline a six-step framework for preparing for AI-driven code modernization, highlighting that as AI accelerates code changes, organizational processes become key bottlenecks.
CUA-SWE introduces a benchmark, environment, and evaluation pipeline that unifies computer-use agents and coding agents, testing whether frontier agents can complete software engineering tasks requiring visual feedback from running applications, across four SWE domains with deterministic tests.
Kent C. Dodds gives a talk titled 'The Last Software Engineer' at Nerdearla 2026, discussing the future of software development in the AI era.
The author argues that coding agents, despite their capabilities, make software engineering more challenging and require extraordinary discipline, referencing recent AI models like Opus 4.6 and GPT-5.2 in the discussion.
Taste-Bench is a benchmark that evaluates LLMs' ability to choose optimal paths at decision forks in long-horizon tasks, using trajectories from software engineering and machine learning research, with a leaderboard indicating current top models like GPT-5.6 Sol achieving 59.7% accuracy.
Simon Willison notes that coding agents, while powerful, add complexity to software engineering, requiring significant discipline and knowledge to unlock their full potential.
A reflection on software engineering emphasizing product judgment over code typing, based on experience at VS Code and a conversation with Kent C. Dodds about AI's role in product engineering.
Kent C. Dodds demonstrates a technique to prevent low-quality code generation without needing to manually review the code.
This paper introduces PLLM+, a hybrid pipeline that combines deterministic replay of historical dependency configurations with LLM-based repair to resolve Python dependency conflicts, showing improved success rates and reduced runtime on the HG2.9K benchmark.
Stanford University's CS329Z course on Engineering AI Agents has started, with lecture slides now public, covering topics from models to autonomous systems and featuring instructors like Diyi Yang and Michael Ryan.
This article discusses the importance of evaluations in AI systems, explains why traditional testing is insufficient, introduces three main types of evaluations, and provides implementation suggestions.