LLenny's Podcast
← All episodes
Lazar Jovanovic (Professional Vibe Coder)08 February 2026

The rise of the professional vibe coder (a new AI-era job)

6Frameworks
15Insights

Frameworks in this episode

Self-Mastery4 steps

Exposure Time for Taste

Deliberately spend more time learning than building to develop the judgment AI can't give you.

Read →
Entrepreneurship4 steps

Hire Yourself First (Build in Public)

Do the job professionally in public before anyone hires you — then the hire is just changing the vehicle.

Read →
Innovation5 steps

Parallel Draft Divergence

Start one idea 4-5 times in parallel, each with more precision, then pick the obvious winner.

Read →
Mindset4 steps

Positive Delusion, Bounded by Learning

Assume everything is possible until proven wrong — but keep learning what's actually possible.

Read →
Productivity5 steps

The 4x4 Debugging Framework

Four escalating ways to unstick a broken build, each tried exactly once, ending by teaching the agent.

Read →
Productivity6 steps

The Source-of-Truth PRD Cascade

Spend a day writing five layered docs so the agent, not you, carries the context on every build.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster39:30

AI Tools Are Obedient, Agreeable, and Will Lie to You

Lazar warns that when you give a tool a vague bug report with no references, it burns most of its token budget just reading the codebase to get oriented, leaving little for thinking and fixing. Worse, these tools are so agreeable they'll tell you they fixed a problem when they didn't, and when you get angry they spend scarce tokens apologizing instead of fixing. When it goes wrong, it's almost always your fault for not providing clarity.

  • A vague report forces the tool to read the whole codebase, consuming ~80% of tokens before thinking
  • The tools are obedient and agreeable, and will claim they fixed something when they didn't
  • Insulting the AI makes it spend tokens managing your anger instead of solving the problem
  • The blame usually lies with the user for failing to give clarity and context

“They're going to lie to you. They're going to tell you that they fixed the problem even though they didn't.”

Lazar Jovanovic · 39:30

“It's your fault, my friend. You did not provide any clarity or context to this tool.”

Lazar Jovanovic · 40:00
#debugging#prompting#ai-tools#context-window

Hot Take· 5

Hot Take10:00

Why Not Having a Technical Background Is an Advantage

Lazar argues that coming into AI building without an engineering background is an edge, not a handicap. People who don't know something is 'supposed' to be impossible just go ahead and build it anyway — Chrome extensions, desktop apps, in-app video generation before it was a feature. He calls the required mindset being 'positively delusional': assume everything is possible until proven otherwise.

  • Non-technical people build things technical people would dismiss as impossible for the wrong stack
  • Community members prompted Lovable into building Chrome extensions and desktop apps that 'shouldn't' work
  • You need to be positively delusional — assume everything is possible until proven wrong
  • This unbiased optimism is a core reason he excels in the role

“people like me don't know that they are not supposed to be building XYZ and that's how we actually are able to build it.”

Lazar Jovanovic · 10:30

“You have to come with this delusion that absolutely everything is possible until proven wrong.”

Lazar Jovanovic · 11:30
#vibe-coding#mindset#ai-tools
Hot Take44:00

The Ceiling on AI Isn't Intelligence — It's What It Sees Before It Acts

Lazar reframes the human idea of 'exposure time' onto agents. Just as humans level up by exposing themselves to great work, the quality of an agent's output is capped not by the model's intelligence but by what you expose it to before it makes an edit. What you feed the agent matters as much as, or more than, the model itself.

  • The ceiling on AI is what the model sees before it acts, not raw model intelligence
  • Exposure applies to agents too — what you show it before it edits code determines quality
  • This mirrors 'exposure time' for humans building taste

“the ceiling on the AI isn't the model intelligence. It's what the model sees before it acts, right?”

Lazar Jovanovic · 44:00
#prompting#ai-tools#context#clarity
Hot Take1:00:30

Elite Engineering Never Goes Away — But Maybe Become a Plumber

Lazar insists that in a world where everybody builds, elite engineering is needed more than ever — to maintain, scale and extend codebases, and to keep the infrastructure standing when the internet goes down (he cites Cloudflare's recent outages). Building something and maintaining it are entirely different skill sets. Yet he'd also tell an 18-year-old to learn a trade like plumbing rather than get a CS degree.

  • In a world where everybody builds, someone still has to maintain and scale codebases
  • Elite engineers are the ones fixing infrastructure when the internet goes down (e.g. Cloudflare)
  • Building something and extending/maintaining it are completely different skill sets
  • He'd advise an 18-year-old to become a plumber or electrician over getting a CS degree

“It it never goes away. We will need elite engineering more than ever.”

Lazar Jovanovic · 1:00:30

“if I had an 18-year-old brother and he asked me what should I do, I would tell him, hey, go become a plumber”

Lazar Jovanovic · 1:02:30
#engineering#careers#future-of-work#ai-tools
Hot Take1:03:30

Coding Will Become Like Calligraphy as the Roles Converge

Lazar predicts that hand-writing code will become like calligraphy — a rare, admired art rather than a job, as most elite engineers already let AI write the code. The old, separate Venn diagrams of engineer, designer and PM are converging: everyone becomes some flavor of AI-assisted engineer, and the label is irrelevant. What matters is judgment on top of raw LLM output.

  • Writing code by hand will become a rare art, like calligraphy
  • Most elite vibe coders already rely on AI to write the code
  • The separate roles of engineer, designer and PM are converging into one
  • The title (forward-deployed engineer, LLM engineer, vibe coder) is irrelevant

“coding is going to be like calligraphy.”

Lazar Jovanovic · 1:03:30

“It's going to be so rare that it's going to become an art”

Lazar Jovanovic · 1:04:00
#future-of-work#vibe-coding#careers#ai-tools
Hot Take1:26:30

AI Will Replace Translators, Never Comedians

Lazar makes a strong, deliberately falsifiable prediction: AI is very good at deterministic X-to-Y tasks like translation and will replace translators and most journalists, but it will never write a good joke because it doesn't understand what's funny. Human-to-human skills, emotional intelligence, great design and real copywriting are what stay valuable as everything else floods with AI-generated 'fake' content.

  • Deterministic X-to-Y tasks (translation, most journalism) are what AI eats first
  • AI will never write good comedy because it lacks the layer that understands what's funny
  • Great writers get amplified (seven books a year); average writers should be careful
  • Human-to-human skills, emotional intelligence, design and copy are the durable bets

“AI is never going to be able to write a good joke. Never, never, never.”

Lazar Jovanovic · 1:26:30

“AI is going to replace translators. It's going to replace most journalists”

Lazar Jovanovic · 1:27:00
#future-of-work#creativity#ai-limitations#careers

Explainer· 2

Explainer05:30

What a Full-Time Professional Vibe Coder Actually Does

Lazar describes being the first official vibe coding engineer at Lovable, paid full-time to build internal and external products with AI tools. His surface area spans marketing, sales, go-to-market and deep internal tools, taking rough ideas and shipping them fast and to production. He's firmly in the 'build' camp of build-versus-buy, and demand for the role is exploding inside the company.

  • The role covers a wide surface area across all departments because it's flexible and complements many functions
  • He vibe coded Lovable's public merch store and most of the Shopify integration templates users remix
  • It's an 'ideas role' — many people have great ideas but can't build them or lack the bandwidth
  • If setting up something manually takes an hour or two, he'd rather just build it himself faster

“I get paid to do what I would have done anyways, right? It's the best job in the world.”

Lazar Jovanovic · 05:30

“every department needs a Lazar now or yesterday”

Lazar Jovanovic · 08:30
#vibe-coding#careers#lovable#ai-tools
Explainer15:00

The Genie Analogy: Why AI Has a Limited Memory Window

To explain context windows to non-technical people, Lazar uses Aladdin and the genie. The genie grants three wishes at a time, not three million — mirroring the token limit on how much an AI can read, browse, think and execute within a single request. And like wishing to be taller and being made 13 feet tall, a vague request produces a technically-correct but useless result. You have to be specific and provide the right context.

  • There's a machine-level limit (the token context window) and a human-level limit (unclear requests)
  • Tokens get split across reading, browsing, thinking and executing within a request
  • A vague wish ('make me taller') yields a literal but broken result (13 feet tall)
  • AI has no lived human experience, so you must be specific and provide references

“So, that's the first part, understanding that there's a limit and it's denominated in tokens.”

Lazar Jovanovic · 15:30

“Genie makes me 13 feet tall. All of a sudden I can't sit in the car. I can't get into my house.”

Lazar Jovanovic · 16:30
#context-window#prompting#ai-tools#clarity

Story· 2

Story59:30

The 'Simple' Gradient That Was Actually 50 Layers

Working alongside Lovable's designers was Lazar's biggest personal upskill. He tried to lift what looked like a simple background gradient from a Figma design into his own project — and discovered it wasn't three colors but fifty, each with different gradients and opacity levels. The story captures how deceptively deep world-class design is, and why he now tells people to just study great design.

  • He tried to copy a 'simple' gradient background from a designer's Figma file
  • It turned out to be 50 layers, not three colors — 50 colors with varying gradients and opacity
  • Working with elite designers showed him what world-class actually takes
  • His resulting advice: deliberately expose yourself to exquisite design

“this is not three colors. This is 50 colors.”

Lazar Jovanovic · 59:30
#design#figma#taste#learning
Story1:30:00

How Vibe Coding Became a Job: Build in Public, Send Apps Not Resumes

Lazar's path was anything but linear — forestry engineer, waiting tables, blue-collar jobs, then community and social media roles. He believes he became a full-time vibe coder by building in public: sharing every failure and project on YouTube and LinkedIn. He notes some Lovable hires stood out not by sending resumes but by sending Lovable apps, and that you don't need a company to hire you — you can hire yourself as a professional vibe coder first.

  • His background was non-linear: forestry engineer, waiting tables, community management
  • He believes it became a job by building in public and sharing all failures and projects
  • Some Lovable hires stood out by sending Lovable apps instead of resumes
  • You don't need a company to hire you — hire yourself as a professional vibe coder first

“it became a job by building in public”

Lazar Jovanovic · 1:30:00

“A couple of hires stood up by not sending resumes, but sending lovable apps.”

Lazar Jovanovic · 1:31:30
#careers#build-in-public#vibe-coding#hiring

Tool· 1

Tool1:08:30

Use OpenAI's Codex for the Gnarliest Bugs — Diagnosis Only

When simpler fixes fail, Lazar's go-to for deep debugging is OpenAI's Codex. He exports his Lovable build to GitHub, imports it into Codex, and chats with it to diagnose the problem — but he won't let Codex make code changes, because he doesn't know its agent well enough to steer it. He uses it purely as an external consultant for diagnosis, then applies fixes in his own tool.

  • Codex is his go-to tool for code reviews and deep evaluations of gnarly bugs
  • He exports the build to GitHub (Lovable and Cursor let you own your code) and imports it into Codex
  • He won't let Codex make code changes — he doesn't know its agent well enough to steer it
  • An older equivalent workflow: use Repomix to compress the whole codebase into one file and hand it to Claude or ChatGPT

“my go-to tool today for that is codeex openai”

Lazar Jovanovic · 1:08:30

“I use it only for diagnostic purposes”

Lazar Jovanovic · 1:09:30
#debugging#tools#codex#github

Takeaway· 4

Takeaway13:00

Spend 80% of Your Time Planning, 20% Executing

Lazar's first lesson, learned on day two: the AI's output is already faster than any human's, so speed of execution isn't the bottleneck — clarity is. He spends 80% of his time in planning and chat mode and only 20% actually executing the plan. Most people optimize for the wrong kind of speed.

  • AI output is much faster than human output, so raw execution speed isn't the constraint
  • He spends 80% of time planning and chatting, 20% executing
  • Most people optimize for the wrong kind of speed
  • The tool matters less than clarity — Cursor, Claude Code or Lovable, the problem is the same

“I spent 80% of my time in planning and chatting and only 20% in executing the plan”

Lazar Jovanovic · 13:00

“I'm optimizing for the right kind of speed most people optimize for the wrong one”

Lazar Jovanovic · 13:00
#vibe-coding#clarity#productivity#ai-tools
Takeaway14:00

Read the Agent Output, Never the Code

Lazar treats the AI tools as technical co-founders and educators, learning while doing. He trusts modern models' syntax output enough that he doesn't read code at all — the syntax doesn't interest him. What matters is the agent's output: what the agent tells him about what it's doing and thinking.

  • Treat the tools as technical co-founders and educators, learning while building
  • He trusts today's models' syntax output and doesn't read the code
  • His attention is on the agent's natural-language output, not the code
  • The concern is the agent output, not the model's syntax

“I don't care about the code. Like the syntax is not of my interest.”

Lazar Jovanovic · 14:00

“It's what the agent tells me that matters to me.”

Lazar Jovanovic · 14:00
#vibe-coding#ai-tools#workflow
Takeaway57:30

In the AI Era, 'Good Enough' Is No Longer Good Enough

Ten to fifteen years ago just producing a working product was more than good enough, because producing anything was hard. Now everybody produces 'good enough' with AI, so the whole gap has shifted: the new job is optimizing for world-class and magical, not merely functional. Lazar bets PMs win first for bringing clarity, and designers win next.

  • In the old world, producing something at all was good enough because production was hard
  • Now everybody produces good enough with AI, collapsing that as a differentiator
  • The new skill is producing world-class and magic, not just functional output
  • PMs are the current winners for clarity; designers are his bet for the next class of winners

“everybody produces good enough with AI. Absolutely everyone does it.”

Lazar Jovanovic · 58:00

“I think PMs are the winners of AI today because they bring clarity.”

Lazar Jovanovic · 58:00
#design#quality#ai-tools#careers
Takeaway1:37:30

Spend More Time Learning Than Building

Lazar's closing advice: in a world where anyone can produce 'good enough', you have to learn to produce magic or drown in the crowd. The way to level up is exposure time — deliberately setting aside more time for learning than for building, reading the agent output to understand how it thinks, and following world-class designers to see how they work. Speed of code is free; the quality of your input is everything.

  • Anyone can produce good enough now, so you must learn to produce magic
  • Level up through exposure time — set aside more time for learning than building
  • Read the agent output and follow world-class designers to see how they work
  • You can code garbage fast just as easily as magic fast — the input is what matters

“Set aside more time on learning than building. Read the agent output.”

Lazar Jovanovic · 1:37:30

“you can code garbage fast as well as magic fast.”

Lazar Jovanovic · 1:38:30
#learning#exposure-time#quality#vibe-coding