Berthd
Berthd introduces a macOS app designed to run AI coding agents on remote 'boxes,' promising continuous operation even when your laptop is closed. However, the Hacker News community largely views this as yet another 'reskin' in a saturated market of AI coding tools, prompting discussions on the lack of innovation and the evolving relevance of human developers. Users are weary of the perceived 'AI product sameness' and question the true value proposition of such offerings.
The Lowdown
Berthd is an open-source macOS application that aims to enhance the development workflow by allowing users to run AI coding agents, such as Claude Code and Codex, on remote 'boxes' or development environments. The primary benefit highlighted is the ability for these agents to continue their tasks uninterrupted even after the user's laptop is closed, with all progress and changes becoming available upon reopening.
Key features and functionalities of Berthd include:
- Persistent Agent Execution: AI agents operate on external 'boxes,' ensuring that coding tasks proceed continuously, independent of the local machine's status.
- Simplified Workflow: The application facilitates the quick setup of a worktree, assignment of a box, and initiation of an AI agent, complete with its development server integrated into a dedicated tab.
- Granular Progress Tracking: Users are presented with a detailed, real-time view of agent activity, including commands, reads, and diffs for code modifications, allowing for in-place interaction and easy publishing.
- Dual Worktree Management: The interface supports managing two worktrees concurrently, with a split-screen view for comparing chats, diffs, and terminals.
- Personalized Interface: Berthd offers extensive theme customization, including nineteen ports of popular styles, chat backgrounds, and scalable text sizes.
In essence, Berthd endeavors to provide a dedicated, always-on, and visually cohesive platform for developers to interact with and manage AI coding assistants, thereby decoupling complex coding tasks from the local machine's uptime and offering a more integrated AI-driven development experience.
The Gossip
Monotonous Moats: The AI Product Proliferation Problem
The most dominant sentiment among commenters was a profound sense of fatigue and skepticism regarding the endless stream of AI coding agent products. Many argued that Berthd, like numerous others, offers little genuine innovation beyond reskinning existing large language models or replicating functionality already achievable with basic command-line tools or established open-source projects like Paseo and tmux. This led to a broader critique that the AI product space lacks creativity, with everyone building 'the same shit' and questioning the long-term viability or 'moat' of such ventures.
Persistent Problems: The 'Close the Laptop' Conundrum
Berthd's flagship feature—agents continuing work after the laptop is closed—was met with a collective shrug and suggestions for far simpler alternatives. Commenters quickly pointed out that merely using `sudo pmset disablesleep 1` or configuring remote VMs (which many developers already do) achieves the exact same persistence without needing a new application. There was also a demand for clearer definitions, asking what exactly constitutes a 'box' and whether it's a local or remote virtual machine, indicating a perceived lack of clarity in the product's fundamental premise.
Code Review Controversy: Humans vs. AI Output
A lively tangent emerged around the practice of reviewing AI-generated code. One perspective provocatively claimed that 'self-review is an anti-pattern' for AI output, likening it to not reviewing compiler-generated assembly—if it works, ship it. This was vehemently challenged by others who highlighted the fundamental difference between deterministic compiler output and the non-deterministic nature of LLMs. Critics emphasized that human review ensures quality, maintainability, and alignment with project goals beyond mere functionality, underscoring that 'it was never about the code' but about understanding the problem space and execution.
Marketing Muddle: Buzzwords and Blurry Value
Several users expressed frustration with Berthd's marketing language, deeming phrases like 'One line. A worktree, a box, an agent.' to be unhelpful buzzwords that obscure the product's actual function and value proposition. This criticism extended to a broader discontent with what they termed 'vibe coding'—building and releasing products quickly without thorough market research or clear problem-solving intent. Some commenters even questioned the authenticity of the story's high upvote count on Hacker News, hinting at potential artificial boosting in a crowded market.