ROUTE, an open-source MCP server I'm building to give AI agents an opportunity layer , from discovering opportunities to eventually understanding, preparing, and acting on them.
Still building out the core and figuring out what the developer experience should look like.
Also trying to build this while getting closer to exam season 😭
I have two tests next week, but I really can’t stop building right now.
I’m just working on ROUTE and using Drift to schedule myself and stay productive.
Trying to balance both without losing my mind.
First time building a browser extension.
It’s been a completely different kind of engineering experience.
I’m building something around a problem I face:
You can have your entire day planned out, but when it’s time to actually work, your browser is rarely ready for the task.
2 weeks into learning Rust, and one thing has surprised me:
My TypeScript background actually helps a lot.
Types, generics, enums, async/await, modules, functions with explicit contracts…
The syntax is different, but the mental model feels familiar.
Implemented my first authentication + payment system for a platform with 1000+ users.
A reminder of how far I've come, from building simple projects to building systems people will actually rely on.
So… I’m leading a dev team now.
First time doing this.
And I’m already learning that leading ≠ just writing code.
Still learning. Still figuring things out.
Let’s see where this goes.
Calendars plan your time.
Task managers organize your work.
Yet focus remains a problem.
Productivity isn't just about planning.
It's about the environment.
Working on an AI agent architecture where the LLM isn't directly coupled to the browser.
The flow is roughly:
LLM → Strands Agent → Tools → Backend → WebSocket → Extension → Browser
The agent handles reasoning and tool selection, the extension owns browser execution.
AI has made writing code easier.
So I'm paying more attention to:
System design
AI engineering
Databases
Security
Less:
"Can I build this?"
More:
"Will it scale?"
"Will it survive failure?"
Software engineering is becoming more about systems than syntax.
Two people can ask AI for a project idea, get almost the same answer, and still build completely different products.
The idea is only the starting point.
Architecture.
UX.
Features.
Execution.
AI gives you the starting direction.
Your decisions determine where you end up.
The scariest failure in an LLM system isn't always a `500`.
Sometimes it's a `200`.
Everything looks healthy:
✓ API responded
✓ Low latency
✓ No errors
But the user gets garbage.
In AI , technical success ≠ useful output.
That's a different kind of reliability problem.
What I've been studying.
What I already understand.
Where I got stuck.
What we talked about before.
That's where AI memory + context gets really interesting.