coding agents

GitHub says it discovered 24 vulnerabilities in Android using its open-source AI security agent, describing it as automated vulnerability discovery at scale on real production software. The company published details in a blog post, and the report is drawing attention as a notable demonstration of AI agents doing practical security research on widely used code.

GitHub says its AI agent identified 24 vulnerabilities in Android applications, adding to growing evidence that autonomous coding assistants can take on security research tasks. The finding, reported by Help Net Security, highlights both the potential of AI-driven bug hunting and the questions it raises about verifying machine-discovered flaws. Security teams are watching closely as AI agents move into vulnerability discovery.

Google Research has open-sourced RRSI, a framework allowing AI agents to refine their own evaluation harness while guarding against overfitting. Announced via MarkTechPost, the release lets developers inspect and build on the underlying code. The announcement is drawing attention from AI practitioners interested in agent self-improvement methods that remain reliable rather than gaming their own benchmarks.

JetBrains has introduced Air, described by the company as a system of products built around agentic software development, where AI agents take on coding tasks alongside developers. The announcement is drawing attention from developers debating what it means for the future of IDEs and how JetBrains plans to compete in the fast-moving market for AI-assisted programming tools.

GitHub researchers report that an open source AI security agent they built uncovered 24 vulnerabilities in Android. The company says the tool was used to scan Android code at scale, flagging security flaws automatically. The disclosure highlights a growing trend of using AI agents to find real bugs in widely used software, and the work is being discussed as an example of AI-assisted security research.

Engineering teams using AI coding agents are finding that continuous integration pipelines have become their biggest constraint, according to a report circulating among developers. As agents generate far more code and commits than human programmers, test suites and CI infrastructure struggle to keep up, forcing teams to rethink how they validate machine-written code at scale.

Developers working with AI agents are separating an agent's decision-making logic from the component that generates code or text, a pattern being discussed under the name Jev. The split lets a reasoning model plan while a writing model executes, and people in the field are debating whether this two-brain architecture improves reliability or just adds complexity to agent workflows.

A JavaScript project called Ponytail, published on GitHub by developer Dietrich Gebert, is gaining attention for its unusual approach to AI coding assistants. Its pitch: make your AI agent think like 'the laziest senior dev in the room,' on the principle that the best code is the code you never wrote. The minimalist tool is drawing interest from developers experimenting with how coding agents make decisions.

A JavaScript project called ECC, published by developer affaan-m, is gaining traction among developers. It bills itself as an agent harness performance optimization system, offering skills, instincts, memory, security, and research-first development workflows for AI coding tools including Claude Code, Codex, Opencode, Cursor and others. It is currently trending as one of the most viewed new repositories, reflecting continued interest in tooling that improves how AI coding agents perform.

Volkswagen has brought on UK autonomous driving startup Wayve as a partner for self-driving technology, according to German tech news roundup coverage. The item also flags cost pressure on notebook maker HP, criticism of German rail ticket apps, arrests following a cyberattack on the FBI, and a revamp of AI coding agent PI. The VW-Wayve tie-up is drawing attention as automakers increasingly rely on AI specialists for driver assistance and autonomous systems.

A developer has released a set of marketing-focused skills for Claude Code and other AI agents, covering conversion rate optimisation, copywriting, SEO, analytics and growth engineering. The project, written in JavaScript and hosted on GitHub, is gaining early traction and is being discussed as part of the growing trend of extending AI coding assistants with specialised professional expertise beyond programming tasks.

A developer has launched Pi pod, a tool for running the Pi coding agent in sandboxed environments on your own server. The project, shared on Hacker News, offers a self-hosted alternative to cloud-hosted agent execution, giving developers more control over security, cost and infrastructure. Early discussion is focused on how it compares with managed sandbox options.

A developer has released an open-source model routing tool aimed at coding agents, claiming it reaches performance on par with Astra-level systems. The launch, shared on Hacker News, lets agents pick the best model for each coding task instead of relying on a single provider. Early reaction is focused on whether the performance claims hold up and how it compares with closed alternatives.

Developer obra has published Superpowers, an open-source agentic skills framework and software development methodology for building AI agents, available on GitHub and written in Shell. The project describes itself as a methodology that works, and it is drawing attention among developers exploring structured approaches to agentic coding and AI-assisted software development workflows.

Anthropic has introduced new modding features for Claude Code, its AI coding tool, letting developers customize how the agent behaves in their workflows. The update gives users more control over the assistant's configuration and output. Developers are discussing how the change could make AI coding assistants more adaptable to individual projects and team preferences.

A new open-source project called context-mode, written in TypeScript and hosted on GitHub, targets context window optimization for AI coding agents. It claims to sandbox tool output for a 98% reduction in token usage, persist session memory, and enforce routing across 17 platforms using MCP and hooks. Developers are taking notice as context management becomes a growing pain point for agent workflows.

Discussion is growing around whether AI coding agents could displace Rust from its position as the most admired programming language. Developers are debating whether AI-assisted tools favor more established languages like Python and JavaScript, or whether Rust's safety guarantees remain valuable even when much code is written by machines. The conversation reflects wider uncertainty about how AI will reshape language popularity.

Developers of AI coding agents are moving their tools from cloud and browser-based offerings toward desktop applications, as users voice frustration over restrictive usage limits. Reports indicate the shift aims to give developers more control and reliability, while complaints about capped usage quotas continue to fuel debate in the programming community about pricing and fair access to AI coding assistants.

A discussion thread on Hacker News asks whether anyone is producing genuinely good code with coding agents such as AI assistants. Developers are invited to share real-world experiences on whether AI-generated code meets production standards or still needs heavy human review.

A new open-source tool called Graphene has been introduced, positioning itself as a data analysis toolkit designed to work with coding agents such as AI coding assistants. The project is available on GitHub and is being presented to the developer community for feedback. Early response is modest but engaged, with developers discussing how such tooling could let AI agents analyze datasets more directly within their workflows.

Earendil Works' open-source project pi is gaining traction as a TypeScript toolkit for building AI agents. It bundles a unified API for large language models, an agent loop, a terminal user interface, and a command-line coding agent, letting developers assemble agents without gluing together separate libraries. Interest is concentrated among developers experimenting with coding agents.