I vibecoded a couple of iOS apps & a full SaaS (SEOZilla.ai) over the past six months and the honest answer is: my best coding agent sessions from 3-4 months ago would make great submissions. Excellent debugging, catching poor architecture choices, back-and-forth problem solving.

But lately? I mostly write product specs, make simple architecture decisions, and do QA. The agents just... handle it. Across the board, Opus, Sonnet, Cursor, whatever you're using, the jump in the last 2 months has been wild.

Which raises a genuine question: what is YC actually selecting for with this prompt? The most impressive sessions are probably from people using worse tools or tackling harder problems. The founders who've figured out the best workflows might have the most boring transcripts.

Anyone else finding that their "best" agent sessions are now the least interesting ones?