Very accurate.

I am thinking to improve my harness Avala by removing all the noise produced by the harness adopting a cleaner UI. This is easy because applied strictly MVVM design pattern so my harness is agnostic of the UI framework. I reviewed one of the most popular harness right now and I couldn't believe some files have 12k lines of code. The technical debt is really impressive.

After I built Avala. I noticed by building the harness with Actor Model could fix many issues specifically when the harness is working with sub-agents.

Someone asked me today why I am building my own harness if there are already many good harnesses out there. The reason is simple: I want to solve my own software engineering problems to solve real-world problems. Today, I added two small features that were bothering me: Selected text is now automatically copied to the clipboard. Commands run by Claude are now grouped if there are more than three together, reducing chat noise. Added support for panning and zooming across all generated diagrams.

I have been cooking since 2023. Imagine building web apps using XAML and C# without JavaScript and without WASM, and they can be deployed natively on Windows, Linux, and macOS. #dotnet Preliminary benchmarks (server-rendered pages per second on our test app): 10x Next.js (cached) 7.8x Blazor static SSR 4.3x Leptos (Rust) 2.1x Go The lowest p99 latency of the seven frameworks tested

Claude is fantastic. AI will replace engineers. I was using Claude code. Suddenly Linux stops working the reason? Claude code deleted the entire home folder. I am running Claude code with Opus 5.5 in a laptop with Arch Linux without sensitive data.

As a software engineer, I find it deeply disrespectful when LLMs are used to reverse engineer paid software. I fully support open source, provided it is used ethically. When you pay for software, you are not just buying code you are paying for a solution to your problem. More than that, you are paying for: Ongoing support Bug fixes and updates Long-term maintainability Infrastructure and reliability If you think a company is abusing its position by charging subscription fees, search for an open-source alternative. If you can't find one, it's usually because reproducing that level of software is exceptionally difficult. That is why there are thousands of note-taking apps, but very few alternatives to platforms like Slack or Discord the underlying architectural complexity is enormous. If anyone thinks it's easy: try building the app yourself using "vibe coding," and then try maintaining it over the long term. Let's see how that goes. I gladly pay for software that delivers real value, and I respect the hard work of the developers behind it. #AI

We are in a bull market of technical debt in software development thanks to LLMs. I'm expecting a new ATH to levels I have never seen before

I am building Avala to solve my own software engineering problems. One of the biggest problems I have with LLMs is how they propose a plan and then end up doing something very different; I wanted to see where exactly and how the LLM changed the plan and why. That's why I am building the story of how the plan was created and how it ended.

I read code and I write code everyday. Here is How I am using AI: I write the foundation using design patterns such as: Domain Driven Design. I set rules with architectural tests that the LLM cannot break. I let the LLM write the boilerplate but never the architecture of my systems.

Software engineering is way more important than before. Anyone can ask an LLM to write code, but coding has never been the problem. If you want to become a good software engineer learning fundamentals is not enough. Go deep in any specific field, master it, and after that, you will be able to properly use AI as a tool to amplify your output.

I started building a web framework in 2023 and finally it will be released next month.

I am writing code by hand. Am I outdated?