How Lazar Jovanovic became Lovable’s first professional vibe coder

Feb 8, 2026 · ai · Source ↗

Published 2026-02-08 - Runtime about 103 min - Watch on YouTube

TLDR

Lazar Jovanovic says elite vibe coding is mostly planning: about 80% chat and 20% execution.
He treats parallel prototypes, reference screenshots, and code snippets as the fastest way to force clarity and avoid AI slop.

Key Takeaways

Lazar says non-technical builders can out-ship experts because they approach Lovable with fewer assumptions about what is “possible.”
The best debugging loop starts in chat, then adds console logs, then escalates to Codex for harder failures.
He argues the real leverage is in quality, taste, and judgment, not tech stack or syntax.
Building in public helped him become Lovable’s first official vibe coding engineer.
He believes product, engineering, and design are converging into a single AI-assisted workflow.

Notes

Lazar Jovanovic is a full-time professional vibe coding engineer at Lovable, building both internal tools and customer-facing products without a coding background.
He ships across departments, including marketing templates, sales tools, internal integrations, feature-adoption tracking, and community tools.
He built public-facing Lovable work like Shopify integration templates and the company merch store, including the shirt shown in the episode.
He says a custom stack can make build-versus-buy simpler: if an external setup takes one or two hours, he often builds it himself.
His first major mindset shift was that coding is not the bottleneck; clarity is.
He says he spends about 80% of his time in planning and chat mode, and 20% actually executing.
He relies more on reading agent output than code output, because the agent’s reasoning and mistakes teach him how to steer better.
His clarity workflow starts with a brain dump, then a more specific project, then reference images or animations, then code snippets from sources like 21st.dev or a build.
He often starts four or five parallel prototypes to compare directions and expose better design choices early.
He uses Mobbin and Dribbble for design references, and says code snippets are better than plain English for pixel-perfect results.
When stuck, he follows a 4x4-style debugging flow: ask the tool to fix it, add logging, inspect output, use Codex, then capture the lesson in rules.md.
He says AI has already shifted work from producing code to deciding what to build, and that taste, exposure time, and emotional intelligence will matter more over time.