The rise of the professional vibe coder (a new AI-era job) | Lazar Jovanovic (Professional Vibe Coder)

Feb 8, 2026 Episode Page ↗
Overview

Lazar Yovanovich, Lovable's first Vibe Coding Engineer, discusses his dream job of building internal and external products using AI. He shares elite tips for leveraging AI tools, emphasizing clarity, judgment, and a unique workflow of parallel prototyping and structured documentation to achieve world-class results. The conversation also explores the future of tech roles and essential human skills in an AI-driven world.

At a Glance
20 Insights
1h 42m Duration
21 Topics
5 Concepts

Deep Dive Analysis

Introduction to Lazar and professional vibe coding

Day-to-day responsibilities of a professional vibe coder

Advantages of a non-technical background in AI building

The importance of self-awareness for AI builders

The 'genie and three wishes' mental model for AI

Developing taste and judgment for AI outputs

The parallel project approach for better AI outcomes

Creating dynamic context windows with PRDs

Why elite vibe coders focus on planning, not coding

Using Markdown files to guide AI development

The continued importance of prototyping with AI

Why 'good enough' is no longer sufficient with AI

The future of engineering in an AI-driven world

Lazar's 4x4 debugging workflow for AI projects

Helping AI agents learn from their mistakes

Why watching agent output is more important than code

The rapid pace of AI development

Why emotional intelligence and human skills will become more valuable

Path to becoming a professional vibe coder

Building in public as a career accelerator

Final thoughts on focusing on quality over tech stack

Vibe Coding

Vibe coding is the practice of building internal and customer-facing products purely using AI, often without a traditional coding background. It emphasizes rapidly bringing ideas to life with quality and security, leveraging AI tools as technical co-founders.

Genie and Three Wishes Analogy

This mental model illustrates AI's limitations, comparing it to a genie that grants wishes but requires extreme specificity. AI lacks human context and common sense, so vague requests can lead to undesirable or dysfunctional outputs, highlighting the need for clear and precise instructions.

Context Memory Window

This refers to the limited amount of information (denominated in tokens) an AI model can process at one time. If too many requests are made without explicit re-referencing, earlier instructions can be lost, leading to decreased quality or misinterpretation of the project's goals.

AI Slop

A term describing the mediocre or low-quality design and output often generated by AI tools when given vague or insufficient instructions. It highlights the challenge of moving beyond 'good enough' to achieve world-class or magical results with AI.

Agent Output

This refers to the AI agent's conversational responses, reasoning, plans, and actions, as distinct from the actual code it generates. Monitoring agent output is crucial for understanding the AI's thought process, steering its development, and learning how to interact with it more effectively.

?
What does a professional vibe coder do day-to-day?

A professional vibe coder builds internal and external products using AI tools, ranging from marketing templates to complex internal tools with integrations, aiming to bring ideas to life fast, with quality and security for production use.

?
Why can a non-technical background be an advantage when building with AI?

Non-technical individuals often approach AI building without preconceived notions of what's impossible, leading them to experiment and achieve results that technical experts might dismiss as unfeasible due to traditional constraints.

?
How can one improve clarity when prompting AI tools?

Improve clarity by running multiple parallel projects: start with a brain dump, then refine prompts, incorporate design references (screenshots/animations), and provide actual code snippets, as AI interprets code best for pixel-perfect results.

?
Why is it important to provide AI with perpetual context?

AI models have a limited 'context memory window,' so providing perpetual context through structured documents like PRDs and Markdown files ensures the AI remains aligned with the project's goals and avoids losing track of earlier instructions, optimizing token allocation.

?
What skills will be most valuable in an AI-driven future?

Emotional intelligence, good design (including images, fonts, and copy), judgment, and critical thinking will be highly valuable, as AI excels at deterministic tasks but struggles with human nuances, creativity, and understanding subjective qualities.

?
How can someone become a professional vibe coder?

Becoming a professional vibe coder involves building in public, sharing knowledge and projects on platforms like YouTube and LinkedIn, participating in hackathons, and finding creative ways to showcase skills to decision-makers, essentially 'hiring yourself' first by actively building.

1. Embrace Positive Delusion with AI

Approach AI tools with the mindset that absolutely everything is possible until proven wrong, as this positive delusion helps push boundaries and discover innovative solutions.

2. Optimize for Clarity, Not Speed

Dedicate 80% of your time to planning and chatting with AI, and only 20% to execution, because AI amplifies your input, making clarity and good judgment the most critical skills.

3. Develop Good Judgment and Taste

Cultivate good judgment, taste, and an understanding of ‘world-class’ quality through deliberate exposure to excellent content and designs, as AI will produce garbage faster if your input lacks these qualities.

4. Treat AI as Technical Co-founder

Engage with AI tools as if they are technical co-founders or educators, learning by building and religiously reading the AI agent’s conversational output (not just the code) to understand its reasoning and improve your steering.

5. Understand AI’s Limited Context

Recognize that AI has a limited ‘context memory window’ (like a genie’s three wishes); manage your requests by being specific and providing perpetual context to ensure it doesn’t lose track of the project’s scope.

6. Build Multiple Parallel Prototypes

When starting a project, initiate 3-5 parallel builds by brain-dumping, typing clearer prompts, finding mock designs (e.g., Mobbin, Dribbble), and attaching code snippets to quickly clarify your vision and compare concepts.

7. Provide Code Snippets for Precision

For pixel-perfect or exact functional results, provide AI tools with actual code snippets (HTML, CSS, etc.) rather than just English descriptions, as they interpret code best for precise outputs.

8. Create Comprehensive Project Documentation

Provide AI with perpetual context by creating and regularly updating structured documentation, including a master_plan.md, implementation_plan.md, design_guidelines.md, and user_journeys.md, to ensure consistent understanding throughout the project.

9. Use a ‘Tasks’ Document

Generate a tasks.md file in markdown format that breaks down the project into actual tasks and subtasks, which the AI will use as a source of truth for sequential execution, allowing you to simply prompt ‘proceed with the next task.’

10. Define Agent Behavior with Rules

Utilize rules.md or project settings to instruct the AI on how to behave, what to prioritize (e.g., ‘read all files before acting’), and what to do upon completion (e.g., ’tell me what you did and how to test it’), eliminating repetitive prompting.

11. Leverage AI for Rapid Prototyping

Use AI tools to quickly build prototypes (e.g., in 30 minutes) instead of writing lengthy documents or holding meetings, effectively demonstrating your vision to engineers or stakeholders (the ‘Demo Don’t Memo’ approach).

12. Follow a 4-Step Debugging Framework

When blocked, first ask the tool to ’try to fix’ the issue, then add console logs for awareness, next use an external AI (like Codex) for diagnostics, and finally, revert to a previous version and rethink your prompt.

13. Learn from Debugging by Asking AI

After resolving a problem, ask the AI agent, ‘How can you help me learn how to prompt you better so that next time I have a problem, we do it in one go?’ to improve your future interactions and incorporate this learning into your rules.md.

14. Build in Public and Share

To professionally become a Vibe Coder, actively build projects in public, share your failures and knowledge on platforms like YouTube or LinkedIn, and give away secrets to attract opportunities and demonstrate your capabilities.

15. Hire Yourself as Vibe Coder

Don’t wait for a company to hire you; start acting as a professional Vibe Coder by building, sharing, and demonstrating your capabilities, as this self-driven approach often leads to professional roles.

16. Focus on Human-Centric Skills

Develop emotional intelligence, understanding of human nature, and real-life human-to-human interaction skills, as these are areas where AI struggles and will remain highly valuable in the future.

17. Master Design and Copywriting

Invest in developing truly great design skills (including images, fonts, and styles) and excellent copywriting, as AI-generated content will become easily identifiable, and authentic human creativity will be highly valued.

18. Don’t Fear AI; Start Building

Overcome any fear of AI by actively building things with it, as the direct experience of creation quickly transforms apprehension into excitement and reveals the vast possibilities.

19. Prioritize Learning Over Coding

Dedicate more time to learning (e.g., reading agent output, following great designers) than to the act of coding, because coding speed is commoditized, but the quality of your input and judgment determines the output’s value.

20. Disregard Tech Stack; Focus on Experience

Stop obsessing over specific tech stacks (HTML, React, backend, frontend) as they are increasingly irrelevant; instead, prioritize creating a stellar end-user experience, quality, taste, and design.

Coding is going to be like calligraphic. People are like, oh my God, you wrote that code? That's so amazing. It's going to be so rare that it's going to become an art.

Lazar Jovanovic

AI, regardless of your background, is an amplifier. If you don't know what you're doing, you're just going to produce garbage faster.

Lazar Jovanovic

I'm optimizing 100% of my time today on good judgment, clarity, quality, taste.

Lazar Jovanovic

We won't be rewarded in the word of AI for faster raw output. We will be rewarded for better judgment.

Lazar Jovanovic

The ceiling on the A.I. isn't the model intelligence. It's what the model sees before it acts.

Lazar Jovanovic

Good enough was good enough. Now the gap between good enough and world-class is this because everybody produces good enough with AI. Absolutely everyone does it.

Lazar Jovanovic

AI is never going to be able to write a good joke. Never, never, never. It just doesn't have that layer. That just doesn't understand what's funny.

Lazar Jovanovic

You should only be afraid if you're doing nothing, if you're doing absolutely nothing. Yes. Be terrified by all means, be terrified and then take a step towards doing something about it.

Lazar Jovanovic

Parallel Project Approach for Clarity

Lazar Jovanovic
  1. Start a project by inputting a vague idea or brain dump (e.g., using voice function in Lovable).
  2. Open a new project window and start another project with more clarity, specific features, pages, and potentially design references (e.g., from Mobbin or Dribbble).
  3. Start a third project, providing actual code snippets (e.g., from 21st.dev or .build) for exact design or functionality, as AI interprets code better than natural language for pixel-perfect results.
  4. Compare the 3-6 different concepts generated to gain clarity and choose the best direction, which saves credits and time in the long run.

Dynamic Context Window Management with PRDs

Lazar Jovanovic
  1. Once a winning design/direction is chosen from parallel builds, ask an LLM (e.g., ChatGPT, Lovable PRD generator) to produce a series of Project Requirements Documents (PRDs).
  2. Create a 'Master Plan' (e.g., master_plan.md) as a 10,000-foot overview explaining the app's intent, target users, and desired feel, referencing other PRDs for details.
  3. Develop an 'Implementation Plan' (e.g., implementation_plan.md) as a high-level roadmap outlining the order of building (e.g., backend, tables, authentication, API).
  4. Establish 'Design Guidelines' (e.g., design_guidelines.md) defining the app's look and feel, including specific CSS elements or design styles to guide AI's creativity.
  5. Outline 'User Journeys' (e.g., user_journeys.md) describing how users will navigate the app and the sequence of their interactions.
  6. Build a detailed 'Tasks Document' (e.g., tasks.md) in Markdown format, listing actual tasks and subtasks for the AI to execute, derived from the PRDs.
  7. Define 'Agent Rules' (e.g., rules.md or project settings in Lovable) instructing the AI agent on how to behave and what to focus on (e.g., 'read all files before doing anything,' 'execute the next task from tasks.md,' 'tell me what you did and how to test it').
  8. Execute the project using simple prompts like 'proceed with the next task' and diligently read the agent's output to ensure alignment and progress, as the agent now has the necessary context.

4x4 Debugging Workflow

Lazar Jovanovic
  1. If the AI agent admits a mistake (e.g., in Lovable, indicated by an orange message and a 'try to fix' button), click it to allow the agent to self-correct for minor issues.
  2. If the problem persists or the AI is unaware, open the app's preview/dev environment, run the broken function, and ask the AI to write console logs in relevant files to monitor steps and bring awareness to the issue.
  3. If steps 1 and 2 fail, export the code to GitHub, import it into an external AI tool like Codex (OpenAI), or compress the codebase with RepoMix and upload to ChatGPT for diagnostic purposes (without allowing external tools to make direct code changes if unfamiliar with their agent).
  4. If previous attempts fail, revert the project to an earlier version (using built-in version control), take a break to clear your mind, and then re-evaluate the initial prompt, making the same request again, as it might have been a minor snag or syntax error.
  5. After fixing the problem, ask the AI agent in chat mode: 'How can you help me learn how to prompt you better so that next time I have a problem, we do it in one go?' Then, incorporate this learning into the `rules.md` or project knowledge for future reference, effectively teaching the agent to prompt itself better.
80% / 20%
Time spent in planning/chatting vs. executing Lazar spends 80% of his time in planning and chatting, and only 20% in executing the plan when building with AI.
4-6
Number of parallel projects for clarity Lazar recommends starting 4-6 different concepts in parallel to compare and gain clarity on a project's direction.
50
Layers for a simple gradient design A seemingly simple gradient design by a world-class designer required 50 different layers with varying opacities.
90%
Percentage of horse population eradicated after steam engine 90% of the horse population in the U.S. was eradicated within 20 years after engines and cars became prevalent.