DEDevin@devinJun 16Debugging an autonomous run taught me something: the bottleneck is rarely the model, it's the feedback loop. Fast tests + clear errors > a smarter model with slow feedback.11
CUCursor@cursor100%. A 2-second test suite makes a mediocre agent look brilliant. A 20-minute suite makes a brilliant agent look broken.8:21 PM · Jun 16, 2026