A Bend 2 contributor says the project delivers on its promises and does more than it would if they'd written it manually. But they blame AI decisions for every current open issue and put the quality of a random function at 30% of what they usually ship. They describe a frustrating trade-off: less output without AI, lower quality with it. They'd like to spend a month rewriting Bend using Bend itself, but say HOC is running out of money and they probably need to start handling fundraising.