This article by Sebastian Raschka explores the fundamental architecture of coding agents and agent harnesses. Rather than focusing solely on the raw capabilities of Large Language Models, the author delves into the surrounding software layers—the "harness"—that enable effective software engineering tasks. The piece identifies six critical components: providing live repository context, optimizing prompt shapes for cache reuse, implementing structured tool access, managing context bloat through clipping and summarization, maintaining structured session memory, and utilizing bounded subagents for task delegation. By examining these building blocks, the article illustrates how a well-designed system can significantly enhance the practical utility of both standard and reasoning models in complex coding environments.
Simon Willison explores "vibe coding" - building macOS apps with SwiftUI using large language models like Claude Opus 4.6 and GPT-5.4, without extensive coding knowledge. He successfully created two apps, Bandwidther (network bandwidth monitor) and Gpuer (GPU usage monitor), demonstrating the potential of this approach. The process involved minimal prompting and iterative development, leveraging the LLMs' capabilities for both code generation and feature suggestions.
While acknowledging the need for caution regarding the apps' accuracy, Willison highlights the efficiency and accessibility of building macOS applications in this manner.
While acknowledging the need for caution regarding the apps' accuracy, Willison highlights the efficiency and accessibility of building macOS applications in this manner.
Stripe's "Minions" are AI agents designed to autonomously complete complex coding tasks, from understanding a request to deploying functional code. Unlike traditional AI coding assistants that offer suggestions line-by-line, Minions aim for end-to-end task completion in a single shot. This approach leverages large language models (LLMs) to handle the entire process, including planning, code generation, and testing. The article details Stripe's implementation, focusing on overcoming challenges like long context windows and the need for reliable tooling. The goal is to significantly boost developer productivity by automating repetitive and complex coding tasks.
LLM coding assistance is moving beyond traditional IDE plugins to powerful, terminal-native agents. These agents, like the new open-source **OPENDEV**, operate directly within a developer's workflow – managing code, builds, and deployments with increased autonomy.
OPENDEV tackles key challenges of autonomous AI, like safety and context management, with a unique architecture featuring specialized AI models, separated planning & execution, and efficient memory. It intelligently manages information by prioritizing relevant context and learning from past sessions, preventing errors and "instruction fade."
OPENDEV provides a secure and adaptable foundation for terminal-first system, paving the way for robust and autonomous software engineering.
OPENDEV tackles key challenges of autonomous AI, like safety and context management, with a unique architecture featuring specialized AI models, separated planning & execution, and efficient memory. It intelligently manages information by prioritizing relevant context and learning from past sessions, preventing errors and "instruction fade."
OPENDEV provides a secure and adaptable foundation for terminal-first system, paving the way for robust and autonomous software engineering.
A new ETH Zurich study challenges the common practice of using `AGENTS.md` files with AI coding agents. LLM-generated context files decrease performance (3% lower success rate, +20% steps/costs).Human-written files offer small gains (4% success rate) but also increase costs. Researchers recommend omitting context files unless manually written with non-inferable details (tooling, build commands).They tested this using a new dataset, AGENTbench, with four agents.
Open-source coding agents like OpenCode, Cline, and Aider are solving a huge headache for developersOpen-source coding agents like OpenCode, Cline, and Aider are reshaping the AI dev tools market. And OpenCode's new $10/month tier signals falling LLM costs. These agents act as a layer between developers and LLMs, interpreting tasks, navigating repositories, and coordinating model calls. They offer flexibility, allowing developers to connect their own providers and API keys, and are becoming increasingly popular as a way to manage the economics of running large language models. The emergence of these tools indicates a shift in value towards the agent layer itself, with subscriptions becoming a standard packaging method.
This article details Spotify's approach to building reliable background coding agents, focusing on verification loops and LLM judges to ensure code quality and prevent functional errors. It explores how these feedback mechanisms contribute to predictable and trustworthy automation in large-scale software maintenance.
Tips for setting up a codebase to be more productive with AI coding tools, including automated tests, interactive testing, issue tracking, documentation, and linters/formatters.
AGENTS.md is a simple, open format for guiding coding agents. It's a dedicated, predictable place to provide context and instructions to help AI coding agents work on your project. The document provides an example AGENTS.md file and details on running a local Next.js website associated with the project.
