# 0hmx - for routing the bus at the start this ca...

Channel: #contributor
Source: https://discord.com/channels/1233487248129921135/1534967503107788830
Started: 2026-08-06T16:52:51.546000+00:00
Last activity: 2026-08-07T15:00:23.271000+00:00

## 0hmx — 2026-08-06T16:52:51.058000+00:00

for routing the bus at the start this can be an real pain how to not block off components haa
Attachment: image.png — https://community.tscircuit.com/media/1534967502847606814

## Seve — 2026-08-06T16:53:49.108000+00:00

Yep exactly

## Seve — 2026-08-06T16:54:44.643000+00:00

It may need to know about future traces for penalties etc

## 0hmx — 2026-08-06T17:22:35.007000+00:00

I just asked it to move the components, and it did it really well. Now I'm thinking—can I just throw these kinds of error at it? Maybe we're just not that good at doing things like this, and moving components is exactly the kind of job AI is great at.
Attachment: image.png — https://community.tscircuit.com/media/1534974985175568414

## Seve — 2026-08-06T17:23:01.268000+00:00

yep 100% agree

## Seve — 2026-08-06T17:23:40.325000+00:00

also i think we should be able to analyze those situations and basically be like "the bus is unnecessarily blocking a component's escape" so that the LLM can work in a loop correcting it

## Seve — 2026-08-06T17:23:56.722000+00:00

basically we need to be able to generate issues for a partially routed PCB

## Seve — 2026-08-06T17:24:29.166000+00:00

if we keep collecting issues and AI can do partial routing, then it could analyze the routing before it attempts a 2hr route job (my AI is often taking 1-2 hours because it increases effort rather than solving the problem)

## 0hmx — 2026-08-07T08:50:06.657000+00:00

We could try building a lightweight LSP. It might help provide more accurate error information, like showing the exact line where an error occurs. I’m not sure how useful it would be in practice, but we could build a toy version and compare which approach the LLM responds to better.

## 0hmx — 2026-08-07T08:52:03.306000+00:00

Eh now i think its the same as giving line and col number with error message maybe its not that great

## 0hmx — 2026-08-07T08:52:49.201000+00:00

Saying this to think about how to give early errors

## 0hmx — 2026-08-07T08:53:14.109000+00:00

just ignore bad idea i think

## Seve — 2026-08-07T15:00:23.271000+00:00

Yea fwiw i think the really hard part is finding and developing the warnings/errors so we should be sure not to avoid that hard problem