7 lessons from 100 hours of vibes coding
(Spoiler alert: You won’t vibe code the next huge SaaS platform over the weekend as a complete non-coder. However, if you’re persistent and willing to learn, you 𝘤𝘢𝘯 achieve impressive results)

First things first. Let's acknowledge how magical AI coding and development capabilities are. While some corners of the internet are already cynical, the reality is more nuanced and varies by use case and user. What I achieved in 45 days would have taken me 6+ months in the past.

With that said, here’s what I’ve learned. For context, I’ve been building a CRM web app focused on encouraging networking momentum. My tools are AI-assisted code editors like Cursor and Windsurf, along with Perplexity and Claude.

1️⃣ 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹𝘀 𝗼𝗳 𝗰𝗼𝗱𝗶𝗻𝗴, 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲, 𝗮𝗻𝗱 𝗽𝗿𝗼𝗷𝗲𝗰𝘁 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 𝗺𝗮𝘁𝘁𝗲𝗿𝘀. I'm not an expert, but I've tinkered with code since I was 12. I'm comfortable with Python, HTML, CSS, and a bit of JavaScript. Professionally, I’ve worked closely with IT and Dev professionals for 10+ years as a product leader. Without this background, I would have struggled much more.

2️⃣ 𝗧𝗿𝗲𝗮𝘁 𝘁𝗵𝗲 𝗔𝗜 𝗮𝘀 𝘁𝗵𝗼𝘂𝗴𝗵 𝘆𝗼𝘂 𝗮𝗿𝗲 𝗺𝗮𝗻𝗮𝗴𝗶𝗻𝗴 𝗮 𝘄𝗲𝗹𝗹-𝗿𝗼𝘂𝗻𝗱𝗲𝗱 𝘁𝗲𝗮𝗺. 𝗔𝘀𝗸 𝘁𝗵𝗲 𝗔𝗜 𝘁𝗼 𝘁𝗮𝗸𝗲 𝗼𝗻 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗽𝗲𝗿𝘀𝗼𝗻𝗮𝘀 𝗱𝗲𝗽𝗲𝗻𝗱𝗶𝗻𝗴 𝗼𝗻 𝘁𝗵𝗲 𝘀𝗶𝘁𝘂𝗮𝘁𝗶𝗼𝗻. Often, you need a developer, but sometimes you need an architect, a code reviewer, a UI expert, or a helpful educator.

3️⃣ 𝗕𝗲 𝘄𝗶𝗹𝗹𝗶𝗻𝗴 𝘁𝗼 𝗶𝗻𝘃𝗲𝘀𝘁𝗶𝗴𝗮𝘁𝗲 𝗮𝗻𝗱 𝗹𝗲𝗮𝗿𝗻. There are various ways to approach problems and architecture. There are countless tech stack options. Don’t let AI choose blindly. The good news: AI is an excellent learning tool. Use it as an opportunity to learn and grow.

4️⃣ 𝗕𝗲 𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗲 𝘁𝗲𝗰𝗵 𝘀𝘁𝗮𝗰𝗸, 𝗯𝗮𝘀𝗶𝗰 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝗮𝗻𝗱 𝗽𝗿𝗼𝗷𝗲𝗰𝘁 𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 (And choose a tech stack you understand)

5️⃣ 𝗔𝘀𝗸 𝘁𝗵𝗲 𝗔𝗜 𝘁𝗼 𝗽𝗹𝗮𝗻 𝗯𝗲𝗳𝗼𝗿𝗲 𝗶𝘁 𝗰𝗼𝗱𝗲𝘀. Ask it to brainstorm multiple approaches, then make a recommendation. Challenge the AI if something seems off. 𝘉𝘰𝘯𝘶𝘴 𝘵𝘪𝘱: ask it to grade its own recommendations on factors like complexity and maintainability.

6️⃣ 𝗪𝗵𝗲𝗻 𝗽𝗿𝗼𝗯𝗹𝗲𝗺𝘀 𝗮𝗿𝗶𝘀𝗲 (𝗮𝗻𝗱 𝘁𝗵𝗲𝘆 𝘄𝗶𝗹𝗹) 𝗮𝘀𝗸 𝗳𝗼𝗿 𝗿𝗼𝗼𝘁 𝗰𝗮𝘂𝘀𝗲 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀. Ask the AI to deeply consider at least two potential root causes and propose at least two solutions before coding fixes.

7️⃣ 𝗗𝗼𝗰𝘂𝗺𝗲𝗻𝘁, 𝗱𝗼𝗰𝘂𝗺𝗲𝗻𝘁, 𝗱𝗼𝗰𝘂𝗺𝗲𝗻𝘁. 𝗔𝘀𝘀𝘂𝗺𝗲 𝘆𝗼𝘂𝗿 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁 𝗴𝗲𝘁𝘀 𝗮𝗺𝗻𝗲𝘀𝗶𝗮 𝗲𝘃𝗲𝗿𝘆 𝗵𝗼𝘂𝗿. This one is huge. Ask the AI to continually add comments to code and create a separate doc outlining key decisions. This way, when the AI reviews existing code in the future, it can read the comments and understand why decisions were made.