the biggest fumble of my life.
tani v1 was a replacement for windows cortana. voice in, siri-style. hardcoded tasks, carried out on command. we'd already wired up groq so i could use llama models for the question feature.
v2 was supposed to land a couple of months later. it never did — my parents felt my education was moving too slowly in grade 9, so i went from the sri lankan government syllabus to a syllabus i had never touched in my life and wrote igcse in six months.
here's the part i still beat myself up over. i was sat there talking to chatgpt about making v2 entirely ai driven. my thinking was: the terminal can do everything — so what if we just let an ai interpret what the user wants and carry out terminal commands to achieve it? i was going to make it build a website from someone just asking for one.
that is exactly what modern agentic harnesses do.
i didn't know to use reddit or github properly back then, so nobody ever saw it. i think about that a lot.
i was building the right thing. i just never told anyone.