What I take from this experience:
- LLMs can be detrimental to deep thinking and they often lock you into a specific way of thinking too early
- Measure your CSS performance from time to time, it's quite easy to regress there
- We should have more offsites
@yucheng In addition to observing the situation of the agent team, I would like to combine it with linear, but the human-machine task seems to be difficult to manage and relate. You know linear will pay after more than 250 issues. What do you do?