Because reliability. You iterate with the AI on the logic with a visual feedback, and "once you agree with the AI of what you want to build... the code is generated deterministically." That avoids the loop where you let AI generate code, it doesn't work, and "for four days you lost time, you lost money and you're super frustrated."
More from this episode
Alessia explains the distinction. "You need to understand what you want to build before generating the code... you iterate with the AI and you have a visual feedback of what is building, and once you agree with the first logic you can generate the code and the code is generated deterministically. So once you agree with the AI of what you want to build, it works, and this is so much faster, so much more robust." The failure mode they avoid: "let the AI generate the code not understanding it, test a bit, it doesn't work, you go in a loop like this for four days, you lost time, you lost money and you're super frustrated." Elia adds a developer's instinct: "when you're a developer you first think a lot before coding, and AI doesn't do that right now... AI just removed everything, like remove the full folder and just redo it again," whereas "the rules in coding is that you have to be very minimalist and you want to change only what is necessary." They focus on the backend — "a very hard problem that a lot of people are not tackling today" — and on deterministic workflows like "connect multiple apps," because "every time that the user connect to your platform you send them an email, you want this to always happen," not have an AI decide whether to.