Lenny's PodcastVibe coder Lazar Jovanovic: How to plan before AI ships slop
How Lovable's first vibe coder spends 80% planning and 20% executing; he runs parallel prototypes and uses sources-of-truth docs to beat context limits.
CHAPTERS
What a “professional vibe coder” does at Lovable
Lazar explains his day-to-day responsibilities as Lovable’s first official “vibe coding engineer,” shipping both internal tools and public-facing experiences. He frames the role as an “ideas-to-production” accelerator that helps any department turn concepts into real, secure, usable software quickly.
Real examples shipped: Shopify templates, merch store, and internal analytics tools
They dig into concrete outputs Lazar has delivered, from customer-facing templates to internal tracking systems. The examples illustrate how vibe coding spans marketing proof points, integrations, and custom internal tooling where “build vs buy” tilts toward building.
Why not having an engineering background can be an advantage
Lazar argues that non-technical builders aren’t constrained by assumptions about what’s “possible,” which leads to creative breakthroughs. He shares community anecdotes where people prompted their way into capabilities that seemed out of scope (extensions, desktop apps, video generation).
The real bottleneck: clarity (not coding) + the 80/20 planning rule
Lazar reframes success as spending most time on planning and clarity rather than execution. He claims AI makes raw output cheap, so the winning strategy is optimizing prompts, decisions, and direction—then letting the agent implement.
Limits of AI today: context windows + the “Genie wish” specificity problem
Using the Aladdin/Genie analogy, Lazar explains two constraints: machine limits (token/context window) and human limits (vague requests). The chapter focuses on why “you know what I mean” fails with AI, and how specificity is foundational to good results.
A practical clarity hack: build 3–5 versions in parallel to find the winner
Lazar’s signature workflow is to prototype multiple directions simultaneously—brain dump, refined prompt, reference screenshots, and even code snippets—then choose the best. This accelerates learning, improves taste calibration, and reduces wasted iteration costs later.
Dynamic context management with PRDs, Markdown plans, and agent rules
To avoid context decay, Lazar externalizes the project’s memory into documents the agent can reread. He creates a small set of PRDs (source-of-truth docs), a tasks.md execution checklist, and an agent/rules file so the tool behaves consistently without repeated prompting.
Why documentation prevents “AI slop” and token waste
They discuss the failure mode where vague bug reports force the agent to scan an ever-growing codebase, wasting tokens on reading and producing low-quality fixes. Lazar argues that missing references cause tools to choose easy-but-wrong fixes, and that emotional prompting (anger) can further waste tokens.
Minimum viable set of files: what to write and what each contains
Lazar outlines the essential documents for someone adopting this workflow gradually. He emphasizes masterplan + implementation plan + design guidelines + user journeys, culminating in tasks.md as the operational driver for the agent.
Debugging without coding: Lazar’s “4x4” unblocking framework
Lazar shares a four-step escalation ladder for bugs and blockers, designed for non-engineers using AI coding tools. The approach moves from built-in “fix” flows to adding observability, then external expert models, and finally reverting and re-prompting thoughtfully.
Compounding learning: turn failures into permanent rules for the agent
After resolving issues, Lazar recommends capturing lessons into rules.md/project knowledge so you don’t rely on memory. This makes the agent progressively more aligned to your preferences and reduces repeat mistakes.
Taste as the differentiator: design, copy, and the shrinking gap to “world-class”
Lazar argues AI makes “good enough” ubiquitous, so competitive advantage moves to taste, design nuance, and emotional resonance. He highlights how pro design often involves hidden complexity (many-layer gradients, typography choices) and encourages studying world-class examples.
The future of roles: engineers, PMs, and designers converge—but elite engineering remains
They explore how AI pushes convergence across PM, design, and engineering as everyone can “build.” Lazar predicts manual code-writing becomes rare (like calligraphy), but emphasizes ongoing need for elite engineers to maintain infrastructure, scale systems, and keep the world running.
How to become a professional vibe coder: build in public and get hired with apps
Lazar shares his non-linear career path and explains how the role became real through consistent public building and sharing. He recommends demonstrating capability through shipped projects—sometimes even submitting a Lovable app instead of a resume—to stand out to hiring teams.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome