I recently published a longer process note on Zenn:

https://zenn.dev/ai_nikitka93/articles/241dd4fec49360

Here is the short version for DEV.

My name is Nikita Kizevich. I also publish as Mikita Kizevich, AI_Nikitka93 and Nikitka AI. I work across AI-generated photography, AI video, music, and AI-assisted software prototypes.

The core workflow is simple:

1. Start with an idea and a real target user.
2. Describe the product flow and constraints to an AI coding agent.
3. Run the result locally or in a browser.
4. Read the logs and browser behavior.
5. Send concrete failures back into the loop.
6. Package only what is honest to publish: demo, prototype, case study, or production-ready project.

The useful part is not "AI wrote code for me". The useful part is the repeatable loop: idea -> implementation -> runtime check -> correction -> public artifact.

What works best:

• small tasks;
• browser screenshots and real UI checks;
• logs instead of vague "it does not work";
• honest public wording;
• keeping GitHub as a living portfolio, not a fixed list.

What does not work:

• asking for a perfect startup in one prompt;
• publishing generated UI without opening it;
• calling every prototype production-ready;
• hiding where AI was used.

My public GitHub is here:

https://github.com/AI-Nikitka93

I am using DEV as a place to write short practical notes about AI agents, LLM workflows, public demo packaging, Telegram bots, and browser-based checks.