Over the past few days I changed the way the uutils project’s sed program handles data to default from characters to raw bytes. This improves compatibility with GNU sed and also performance. Given the change’s size and extent (13 changed files, 1740 insertions, 609 deletions), I worked with an AI agent (OpenAI Codex), which allowed me to experience first-hand both its power and limitations.
Continue reading "Why agentic AI needs better experts"With ever-more program code created through generative AI, it is important to document its provenance. In industry this can help diagnose the root cause of problems discovered long after the code was written, locate issues with a similar cause, and evaluate the cost and benefits of AI. In educational settings it can help students be honest, mindful, reflective, and transparent regarding their use of generative AI and provide educators with data they can use to evaluate objectively student work. In both settings, provenance information can guide a code review’s focus because humans and AI fail in different ways.
Continue reading "Documenting AI-generated code commits"Here’s the thing: when working on code with GenAI assistance (from a chat-bot, through IDE auto-completion, or, increasingly, with an AI agent) you need a better understanding of the system than when working without. Cognitive psychology and the workings of large language models (LLMs) give us four clues on why this happens.
Continue reading "Why reviewing AI-generated code is devilishly hard"In a keynote presentation at the 2026 Mining Software Repositories Emerson Murphy-Hill, a star researcher at Microsoft, presented his view on the role of an empirical software engineering researcher in the age of generative AI. His talk focused on three themes: the durability, differentiation, and dissemination of research.
Continue reading "Empirical software research in the age of AI"We are living through a golden age of generative AI: a time when progress is both breathtaking and somehow still unable to reliably open a PDF without hallucinating the author’s middle name. The curve is real, the funding is real, and the demos are so real that they must be watched on a stage with dramatic lighting, because ordinary lighting reveals too much. The modern model does many things remarkably well—summarization, translation, code generation—and then it will confidently assert that 9 is a prime number “depending on your threat model,” which is how you know you’re witnessing history.
Continue reading "Vibe coding toward the incident horizon"
Last update: Wednesday, July 8, 2026 1:30 pm
Unless otherwise expressly stated, all original material on this page created by Diomidis Spinellis is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
