Lazar Jovanovic on Vibe Coding as a Profession, the Genie Framework, and PRD-Driven Development

Speaker:
Lazar Jovanovic
Source:
Lazar Jovanovic on Vibe Coding as a Profession, the Genie Framework, and PRD-Driven Development · ~Early 2026
Episode:
Listen →
transcriptlennys-podcastvibe-codingai-toolsproduct-managementsoftware-developmentlovabledesign

Lazar Jovanovic on Vibe Coding as a Profession, the Genie Framework, and PRD-Driven Development

Key ideas

  • Clarity, not coding, is the bottleneck. Jovanovic has no software engineering background. His competitive advantage as a professional vibe coder is not technical — it is clarity: knowing precisely what to build, what good looks like, and how to communicate intent to an AI agent. The genie analogy captures this: you rub the lamp, a genie appears, and you ask to “be taller.” The genie makes you 13 feet tall because you were not specific. AI agents face the same constraint — they will grant the wish you describe, not the wish you intended. All skill in vibe coding is skill in specification.
  • Four parallel builds as a clarity engine. Rather than starting with a single prompt and iterating, Jovanovic’s method begins with four simultaneous approaches: (1) a voice/brain dump to get ideas out; (2) a typed prompt with more deliberate structure; (3) a screenshot from Mobbin or Dribbble attached as a visual reference; (4) a code snippet from a library like 21st.dev for pixel-perfect fidelity. Running all four in parallel generates options that make the right direction obvious, and it saves far more tokens in the long run than trying to refine a single wrong direction.
  • PRD-driven development: outsource context to documents. The AI’s context window is finite. Jovanovic’s system beats this constraint by spending a full day creating four planning documents before writing a single line of code: (1) masterplan.md — the 10,000-foot intent and audience; (2) implementation.md — the ordered sequence of technical decisions; (3) design-guidelines.md — exact fonts, colours, and layout rules including CSS where needed; (4) user-journey.md — the full flow for primary and secondary users. A fifth file, tasks.md, generates granular subtasks from these. The agent reads the plan files before every action; the human prompts only “proceed with next task.” Context is never lost.
  • Four-by-four debugging: systematic unblocking without code knowledge. When something breaks: (1) click the agent’s built-in “try to fix” prompt; (2) if that fails, manually add console.log debugging lines to surface the error in browser dev tools, paste the output for the agent; (3) use an external tool (Codex, Repomix + Claude) as an independent diagnostic consultant; (4) revert to a previous version and rewrite the prompt with more precision. Critical: after every successful fix, ask the agent to explain what prompted the mistake and update rules.md so the pattern never recurs.
  • Judgment and taste, not output speed, are the durable skills. As AI models improve, the manual workarounds vibe coders use today — PRDs, task files, parallel builds — will be automated. What will not automate: the taste to recognise whether an output is world-class or merely good-enough, the judgment to know whether something should be built at all, and the emotional intelligence to understand how an interface will make a user feel. Jovanovic explicitly de-prioritises getting better at the workflow mechanics and invests instead in design exposure (following world-class designers on social media), copy quality, and first-principles thinking about user experience.

Overview

Lazar Jovanovic holds the title of “first official vibe coding engineer” at Lovable, the fastest-growing startup in history at time of recording. He has no software engineering background (forestry engineering degree, then community management and startup operations). He discovered Lovable seven months before this recording and built his way into a full-time role by building in public on LinkedIn and YouTube. The episode covers: what professional vibe coding actually looks like day-to-day, the four-parallel-builds method, PRD-driven development (masterplan.md and task.md systems), the four-by-four debugging framework, how to think about context window management, when to use tools like Codex for diagnostics, design and taste as the emerging differentiator, and the career path to vibe coder as a formal role.