yubin hong @studio4any - Twitter Profile

A human reviewer tops out around 400 lines of code per hour before quality degrades. AI now writes far more than that, far faster. The bottleneck moved from writing code to reading it, and most teams haven't noticed yet 👇 https://t.co/7IcxbY30e8

4

0

171

yubin hong

@studio4any

4 days ago

@BrassCoder thanks for reading!

0

2

yubin hong

@studio4any

4 days ago

@X_supporttop1 yes pls!!

0

8

yubin hong

@studio4any

5 days ago

Morning runner

0

12

yubin hong

@studio4any

6 days ago

The API boundary / Routing layer, for sure. Pushing 90% to a smaller model sounds clean in theory, but reliably classifying task difficulty before running the call without adding latency or cost is where people trip up. Most end up over-routing to frontier models just to avoid unexpected breakages

0

3

yubin hong

@studio4any

7 days ago

Your coding agent is three things wearing one costume. The harness. The inference provider. The model. Most people never separate them, so they pay frontier prices for work a small model could do. I spent this morning taking it apart 👇 https://t.co/A4q0oPSQbh

6

3

0

1

237