Tag
The article discusses emerging security concerns with AI coding agents, focusing on permission management and the risks of giving them full access to development environments.
Yoetz is a local-first, open-source work ledger for AI coding agents that verifies actions and provides receipts of verified work, supporting Codex, Claude, and Cursor.
This article compares the performance of four AI coding agents—Cursor, Claude Code, Codex, and Antigravity—by testing them on the same app build and a broken production app with hidden bugs.
The author describes challenges in team collaboration with AI coding agents, such as conflicting decisions and context drift, and seeks advice from the community.
The tweet demonstrates how Cline Desktop, an open-source app for open-weight models, automatically scanned and fixed intentional vulnerabilities in a codebase using a scheduled agent, showcasing AI coding agents as autonomous teammates.
Security researchers discovered critical vulnerabilities in the default GitHub Actions configurations for Anthropic's Claude Code, Google's Gemini CLI, and OpenAI's Codex, allowing remote code execution through unauthenticated issues. A related privilege escalation flaw in Google's ADK repo highlights systemic weaknesses in the CI/CD scaffolding for these AI agents.
The article highlights how coding agents introduce technical debt in software systems and proposes using tools like ArchUnitPython to enforce architectural rules and prevent structural issues.
The Anthropic talk highlights that Claude Code's biggest challenges involve integrating with external workflows and managing limited context windows, emphasizing the need for customization and better feedback in AI agents.
A web developer shares experiences building systems around AI coding agents for reliable development and seeks advice on workflows and orchestration tools.
The author discusses building policy enforcement tools for AI coding agents that include hard allow/deny rules and intent checks to ensure agents follow their stated plans.
Multica is an open-source workspace designed to transform AI coding agents into virtual colleagues that can autonomously accept tasks and report proactively, thereby changing how developers collaborate with AI tools.
The article discusses how AI coding agents can gradually cause issues in codebases and provides rules to minimize damage, applicable across development environments.
The post argues that AI coding agents require more explicit control and rules to prevent unintended behaviors, suggesting a shift towards less freedom rather than more in their development.
A tweet shares that developers using AI prompting for Jira tickets can be 20-30% faster, and highlights a podcast episode featuring the CEO of OpenHandsDev discussing AI coding agents.
The article discusses how AI coding agents enable small software teams to work with the scale of large teams by running multiple agents in parallel, emphasizing the importance of code modularity for effective parallelism.
AI coding agents write runtime adapters to bridge gaps between new model architectures and software stacks, enabling thousands of HuggingFace models to run on the IBM Spyre accelerator with day-one support.
Simon Willison discusses how AI coding agents impact software development, including productivity metrics like lines of code and the challenge of maintaining conceptual integrity, using analogies such as the Winchester Mystery House.
Coder Agents is a platform for running AI coding agents on enterprise-controlled infrastructure, offering centralized governance, model flexibility, and auditing capabilities.
This paper identifies critical flaws in current benchmarks for evaluating AI coding agents, such as the misapplication of the pass@k metric, and proposes new metrics like reliability@k and security-adjusted reliability@k to improve measurement of reliability and security.
This describes a set of agent skills for real engineering with coding agents like Claude Code and Codex, aimed at fixing common issues in AI-assisted development.