Vibe Coding & Hands-On Practice
A live demo of AI-built tools, a look at where the world stands on AI adoption, and hands-on prompting practice on real tasks from your own to-do list.
Key Takeaways
- Vibe coding — talking to AI to build apps and websites without touching code — is already practical and within reach for non-developers
- Most people in the world still aren’t using AI at all; being on a paid plan puts you ahead of the vast majority
- Before diving into a feature like Deep Research, step back to delegation: ask what AI can actually do well for your specific task
- Voice dictation is faster than typing for the delegation phase — brain-dump context freely, then refine
- Describe your problem to AI first and let it suggest the best approach, rather than choosing the tool before you know the goal
What We Covered
Vibe Coding in Practice
Otakar opened the session by showing a custom client onboarding flow he built entirely through conversation with Claude Code — no manual coding involved. The tool presents clients with a step-by-step journey: an intro screen, key terms with checkboxes, a date picker, and an invoice form that feeds directly into his CRM. Total build time: around eight hours over one evening and the following morning.
- Zero subscription cost – Unlike Typeform at $25/month, the tool runs on his own domain for free
- Reusable by design – Setting up the next client takes only 2–3 prompts and roughly ten minutes
- Lovable.dev – A beginner-friendly vibe coding platform; useful for landing pages, portfolios, or lightweight apps
“I basically didn’t touch the code. AI coded everything. The whole reason why I’m telling you this is to excite you — you can create your own websites and you don’t need to write code, you can just talk to AI.”
Where the World Actually Stands on AI
Otakar shared a diagram from Steven Bartlett (The Diary of a CEO) showing global AI adoption. Each dot represents 3.2 million people:
- Gray (majority) – People who don’t use AI tools at all
- Green – People using free AI chatbots
- Yellow – People paying ~$20/month for a plan — where most participants in this group sit
- Red (tiny) – People using AI for coding (vibe coding included)
The takeaway: the overwhelm of “everyone is using AI” is misleading. Paying for a tool and using it consistently already places you in the global top tier of AI adoption.
“Pay for an AI tool, use it every day for a month. Read one book or watch a long explainer video. If you already do these three steps, you will be in the top 1%.”
Hands-On Prompting: The Delegation Loop in Action
Participants picked a real task from their to-do list and worked through it live. Otakar coached them using the Delegation–Description–Discernment loop introduced in earlier sessions:
- Delegation – What can AI actually do well here? Define the right sub-task before starting
- Description – Describe the problem clearly; be specific about which file, column, or piece of data you mean
- Discernment – Compare the output to what you actually wanted; decide whether to refine or start fresh
Key coaching point: don’t jump straight to a powerful feature like Deep Research. Start a new chat, describe the problem in plain language, and ask AI how it can help — then let it recommend the approach.
Using Voice Input for Faster Context-Setting
Otakar reminded participants of the microphone icon inside the chat window. During the delegation phase — when you’re still figuring out what you want — speaking is much faster than typing. The AI can handle messy, stream-of-consciousness dictation and still extract the key context. You can always clean up the prompt in the next message once the direction is clearer.
Use Cases
Topics and challenges participants brought to the session.
Questions Asked
Q Is the AI adoption statistic based on real survey data?
The source wasn’t explicitly stated in the LinkedIn post. The diagram includes all 8 billion people on Earth — including children and people in contexts where AI is simply irrelevant. Even if you adjust for that, the proportion of active paid AI users remains very small globally.
Q Will Deep Research work for verifying company names and countries across a large list?
Deep Research is designed for broad topic exploration, not systematic record-by-record verification. For a list of ~100 companies, it’s better to start a new chat, attach the file directly, describe the specific column and what you need, and ask AI how it can help — it will likely suggest a more targeted and reliable approach than deep research.
Q How do I start a new chat inside a project in ChatGPT?
Navigate into the project from the sidebar, then look for the new chat button within that project view. Starting the chat from inside the project means it inherits any files or instructions you’ve stored there, without you having to re-attach them each time.
Homework
- Continue working on the real task you started today — use the Delegation–Description–Discernment loop and bring what you find to the next session
- Try using voice dictation (the microphone icon in the chat) when setting up context for a new task — notice whether it feels faster than typing