We need to be honest about the current state of AI coding tools: we are spending way too much time talking to them.
Don't get me wrong, I love LLMs. But the daily workflow of a developer using AI right now looks like this:
Run the server.
Server crashes with a massive red stack trace.
Highlight the error. Ctrl+C.
Open your AI chat window. Ctrl+V.
Write a prompt explaining the context.
Wait.
Ctrl+C the fix. Switch back to your IDE. Ctrl+V.
You aren't coding anymore. You've become a middleman between your terminal and a chatbot. Every time you context-switch, you lose your flow state. I got so exhausted by this loop that I decided to build a way out.
The Original Idea (And why it wasn't enough)
When I first started building Glass Arc, the goal was simple. I wanted an AI assistant inside VS Code.
But I quickly realized that building just another chat panel wasn't solving the core issue. An assistant waits for you to tell it what to do. It fetches your coffee. I didn't want an assistant; I wanted an architect—a tool that actively watches the workspace and does the heavy lifting before I have to ask.
So, I ripped up the original concept and pivoted. The IDE already knows everything. It knows when you save a file, it knows what your dependencies are, and it definitely knows when your terminal is screaming at you. Why shouldn't the AI just listen to that?
Evolving into an Autonomous Architect
I started wiring Glass Arc to be proactive instead of reactive. I focused on two massive pain points:
The "Ripple Effect" of hitting Save
When you make a structural change to a core file, it usually breaks three other files. I built a system into Glass Arc that maps the "Ripple Effect." When you hit save, the AI quietly scans your workspace, validates side effects, and drafts the necessary cross-file updates to keep your architecture intact. No more chasing broken imports.Auto-Healing the Terminal
This was the holy grail. I wanted Glass Arc to watch the terminal, catch a stack trace, and instantly pop up with the exact file injection to fix it.
I got it working beautifully in my local testing environment. The server would crash, Glass Arc would intercept the red text, read the broken file, and inject the fix. It felt like absolute magic.
And then Microsoft killed it.
The Production Boss Fight
Deploying a VS Code extension is a humbling experience. I packaged Glass Arc, uploaded it to the VS Code Marketplace, and... it instantly crashed on boot for every user.
Why? Because in my local testing, I was using an API (terminalDataWriteEvent) that allowed the extension to secretly read everything typed or outputted in the terminal. Microsoft’s Marketplace heavily restricts this to prevent malware from stealing passwords. The second my published extension tried to listen to the terminal, VS Code executed it.
I had to completely pivot the architecture. Instead of an illegal background listener, I built a compliant, manual interceptor. Now, when a crash happens, you just hit a shortcut, Glass Arc ingests the terminal state, reads your file context, and heals the codebase. It’s completely secure, but retains all the autonomous magic.
(Oh, and my internet dropped literally the second I tried to upload the final .vsix file to the Marketplace. Status Code 0. Because of course it did).
The Way Out
The era of copy-pasting code back and forth to a chatbot is a stepping stone, not the final destination.
Glass Arc is officially live on the VS Code Marketplace today. It’s designed to be the first truly autonomous AI architect that lives in your IDE, maps your dependencies, and heals your runtime crashes.
Live on the VS code Marketplace, please install!: [https://www.glassarc.dev/]
I'd love for you guys to try it out, Let me know what you think in the comments!
Top comments (0)