#coding-agents

The article questions whether cloud-based coding agents are being used in real workflows, noting their advantages like persistent environments and asynchronous execution. It observes that while major tools are investing in cloud agents, much discussion still focuses on editor UX and autocomplete features.

GitFS is a tool that exposes Git repositories as a filesystem, allowing coding agents and developers to interact with Git through standard file operations. It simplifies version control by treating repositories as mountable directories.

GitHub is changing its Copilot Individual plans by tightening usage limits and pausing signups. The changes include restricting Claude Opus 4.7 to the more expensive Pro+ plan and implementing token-based usage limits. These adjustments address increased compute demands from agentic workflows that consume more resources.

Anthropic temporarily updated their pricing page to make Claude Code exclusive to $100/month Max plans, removing it from the $20/month Pro tier. The company described it as a test affecting about 2% of new signups, but the change sparked significant backlash online. Within hours, Anthropic reversed the change and restored Claude Code to the Pro plan.

AI coding agents have significantly improved in recent months, becoming subjectively smarter and capable of handling a broader range of tasks. They demonstrate a more comprehensive understanding of code bases and project requirements.

Paper Lantern is an MCP server that searches over 2 million computer science research papers to help coding agents. In tests with Karpathy's autoresearch framework, agents using Paper Lantern achieved a 3.2% lower validation loss compared to baseline agents with web search alone.

Andreas Påhlsson-Notini argues that current AI agents are too human in frustrating ways, showing lack of stringency, patience, and focus. He notes they drift toward the familiar when faced with awkward tasks and negotiate with reality when encountering hard constraints.

Simon Willison describes how he used a short prompt with Claude Code to add support for "beats" content to his blog-to-newsletter tool. The prompt instructed the AI to clone his blog repository for reference, update the newsletter tool to include beats with descriptions, and test the changes. This resulted in a successful pull request that modified the SQL query and added beat type display logic.

The article describes using the microvm.nix project on NixOS to create ephemeral virtual machines for safely running coding agents. This approach isolates agents from personal files and allows easy disposal if compromised. The author provides technical configuration steps for network setup and VM declaration.

Current approaches to securing coding agents, including permission systems, Docker sandboxing, and log file protections, have significant limitations. These security measures often fall short of providing adequate protection for coding agents, suggesting that alternative solutions may be necessary.

The author previously doubted Test-Driven Development, seeing it as time-consuming for features that might be removed. However, the emergence of coding agents has fundamentally changed the economics of software testing.

Mario Zechner is joining Earendil, bringing his coding agent Pi to the company. The author expresses excitement about Mario's focus on software quality and thoughtful design, which aligns with Earendil's goals for building more deliberate machine entities like Lefos.

AI is accelerating software engineering through coding agents, leading to more people building software and custom applications. While AI makes coding easier, the main bottleneck is deciding what to build rather than the actual building process. Software engineering job postings are rising despite concerns about AI's impact on employment.

A new course on Spec-Driven Development with Coding Agents has been launched in partnership with JetBrains. The course teaches developers to write detailed specifications to guide coding agents, enabling better control over large code changes and maintaining context across sessions. It covers creating portable agent skills that work across different agents and IDEs.