Bilingual

Burned out creating public software
25 days ago
- OpenScreen, an open-source alternative to Screen Studio, was archived due to creator burnout despite 38k stars and feature completeness.
- Managing public software alone leads to overwhelming bug reports and feature requests, highlighting struggles with delegation and maintenance anxiety.
- Proposes AI-driven solutions: one model with a human operator handling features via token donations, or fully autonomous AI agents managing software with minimal oversight.
- Suggests experimenting with current LLM capabilities to test feasibility of AI-managed projects, noting a personal success using AI for rapid feature addition.
Good APIs Age Slowly
23 days ago
- APIs that initially impress often lead to later maintenance troubles; trust in longevity over initial elegance.
- The first version of an API is often overvalued because it's judged in a simple context before real-world dependencies emerge.
- Key API design issues relate to boundaries: deciding what's public vs. private early to avoid unintended dependencies.
- Expose as little as possible initially, as adding later is easier than removing after others rely on it.
- Convenience in APIs can hide assumptions, leading to stiffness with new use cases; boring, honest APIs tend to last longer.
- Avoid designing APIs around current frontend shapes to prevent coupling to temporary UI decisions.
- Versioning doesn't excuse poor API design; stable APIs build trust and reduce coordination costs.
- Good APIs are careful about what they expose, especially in responses, to prevent unintended reliance.
Pros and Cons of Solo Development
21 days ago
- Solo development allows for complete freedom, including designing the app exactly as desired, making all product decisions, and choosing licensing.
- It involves full responsibility for everything, requiring significant self-discipline and time, with challenges in maintaining code quality, lacking peer feedback, and handling support.
- Learning opportunities are abundant, covering aspects beyond typical job niches, but balancing pros and cons is key to enjoying the process.
- The project was started due to personal dissatisfaction with existing tools, inspired by an abandoned app, and has grown to a community with over 350 GitHub stars.
- AI aids in coding but adds complexity in managing large contexts and preventing poor quality, while no telemetry limits understanding of user needs and usage patterns.
- The experience is rewarding, especially when users appreciate the app globally, making the effort worthwhile despite the sacrifices.
A new study just debunked the biggest fear about AI and open source
20 days ago
- A study on AI coding agents in GitHub repositories found no significant decline in newcomer participation, contrary to fears that AI would push beginners out.
- While code complexity increased (e.g., cognitive complexity rose about 11% in Python projects), it did not deter new contributors, and the active contributor base grew.
- The research had limitations, including focusing on established projects with pre-AI history and measuring adoption via configuration files rather than actual tool usage.
- Despite AI-generated code being more complex, newcomer retention held steady, and the main concern shifts to the growing maintenance burden on existing maintainers.
- The study suggests future work should explore AI reliance levels and projects 'born with AI,' emphasizing the impact on maintenance efforts over contributor numbers.
We charge $10k a week to delete AI-generated code
20 days ago
- We analyze AI-generated codebases that have become bloated and difficult to maintain.
- We offer a free analysis and commit to a specific code reduction target (e.g., from 100,000 lines to 35,000).
- The process involves a week of focused work with three senior engineers, starting with a detailed checklist of app functionality.
- We refactor code by consolidating duplicates, replacing custom frameworks with libraries, and rebuilding unsalvageable parts.
- You receive the reduced codebase, QA checklist, guardrails (like CLAUDE.md and lint rules), and a two-week warranty.
- Pricing is $10,000, with payment proportional to the achieved reduction target; only pay for results delivered.
- Lines are counted fairly (non-blank, non-comment), and we avoid code golf to ensure readability.
- We are real engineers (not agents) using Claude Code under strict oversight, based on decades of experience.
- Contact maciej@odra.dev for a free analysis of your repository if this resonates with your project challenges.
Roko's Dancing Basilisk
12 days ago
- Author tests DeepWiki, a tool that generates documentation for GitHub repositories.
- Documentation for mod_blog (7,400 lines) was mostly accurate but had inconsistencies like incorrect layer count and interface issues.
- Documentation for a09 (9,500 lines) contained significant errors, including incorrect backend matrix tables and misrepresentations of features.
- Author attributes errors in a09 documentation to higher cyclomatic complexity and potential LLM context window limitations.
- Concerns raised about updating documentation for evolving codebases and potential inaccuracies in legacy projects.
- Overall assessment: less problematic than LLM-generated code but too unreliable for unfamiliar codebases.

Hasty Briefsbeta

software maintenance

Burned out creating public software

Good APIs Age Slowly

Pros and Cons of Solo Development

A new study just debunked the biggest fear about AI and open source

We charge $10k a week to delete AI-generated code

Roko's Dancing Basilisk