I am working on changing this. For a long time, I’ve used the mindset—"I’m just a beginner; I lack the necessary expertise and perspective"—as a barrier to hold myself back. I haven’t dared to submit pull requests (PRs) because I’m worried about creating extra work for the open-source team, or—worse—contributing "garbage" generated by AI.🥲
It’s great that T3 Code now supports internationalization, but the level of polish is, frankly, a bit "surprising."
But just last weekend, I forked T3 Code and implemented internationalization myself.
Could you perhaps share some details about upcoming release plans? I feel like I’ve wasted my tokens.
No ill intent—just a simple question.
@jullerino@theo@t3dotcodes@maria_rcks ୧₍˄·͈༝·͈˄₎୨Thank you for your patient response. I have instructed my grok bot to create a scheduled task to check merged PRs daily and assess whether they need to be merged into my T3 branch
Sorry, I don't quite understand what you mean by "not good". If you're referring to the translation in T3, I think it's excellent. If you're referring to my previous reply, what I meant is that general translation might misinterpret my original intention or express my question in a way that doesn't quite align with your habits
I am currently learning English. This post was translated by grok for me. If there are any inaccuracies in the expression, please forgive me. Thank you very much
I'm trying to grow from a beginner into a real engineer, and figuring out how to work with AI agents is a big part of that. Would love some input.
I started with "one question, one session." It broke fast. I'd forget to switch, only switch once context was full, re-explain everything, and burn tokens. My sidebar filled up with sessions I couldn't track.
After using T3 Code and digging into how it works, I'm trying this:
- One feature = one PR = one main session
- Main session sets goals + acceptance criteria, never writes code
- Subagents ship one verifiable commit each
- Each change gets an AI self-review, then a cross-check from a model in a different family
- I only review results
T3 Code's inbox sidebar helps a lot. I only see sessions that need me.
Writing acceptance criteria up front cut a lot of rework for me. Cross-model review seems to catch things self-review misses, but I'm not sure it's worth the extra tokens yet.
I put it together as ning-stack, drawing on Matt Pocock's skills, pstack, and Theo's tstack, to fit how I work.
Still figuring it out:
- How do you keep the main session from drifting on long features?
- Is cross-model review worth it, or overkill?
- If I mostly review instead of write, how do I keep actually learning engineering?
- What would you change?