From Vibe Coding to Agentic Engineering: Andrej Karpathy’s Vision for the Future of Software
How AI is shifting software development from manual coding to orchestration and oversight
Welcome to the 39th issue of AI Agents Simplified 🍻
This issue is brought to you by HubSpot
AI coding tools are moving quickly from autocomplete assistants to systems that can plan, execute, debug, and iterate across complex workflows. In a recent talk, Andrej Karpathy outlined a framework for understanding that shift: vibe coding, agentic engineering, and Software 3.0.
His core argument: software development is moving away from writing every line by hand and toward managing intelligent systems that generate and refine code on our behalf.
The change is not just about productivity. It changes what programming is, what skills matter most, and how developers will work with AI systems over the next few years.
Karpathy’s Framing:
Karpathy describes vibe coding as the phase where developers use language models to quickly generate working code through natural language prompts. It lowers the barrier to building, speeds up experimentation, and makes prototyping dramatically easier.
But he argues that this is only the beginning.
The more important shift is toward agentic engineering: using AI systems that do more than generate snippets. These systems can break down tasks, use tools, run tests, recover from errors, and iterate toward an outcome. In this model, the developer’s role shifts from writing everything manually to supervising and directing the process.
Demystify AI for Your Business in Just 4 Steps
Overwhelmed by AI? HubSpot's free guide cuts through the noise. Get the ultimate crash course for non-technical entrepreneurs who want to harness AI's power—without getting lost in the jargon.
"AI for Business Builders" delivers:
A 4-part roadmap to AI mastery
Jargon-free explanations of large language models
Practical prompt engineering tips you can use today
Real-world examples of AI boosting businesses like yours
Arm yourself with the knowledge to make informed AI investments and skyrocket your startup's growth.
❤️ The best way to keep our mission going - at zero cost to you - is by checking out this free guide from our sponsor.
The Rise of Software 3.0:
Karpathy extends this idea with his “Software 3.0” framework:
Software 1.0: traditional hand-written code
Software 2.0: neural networks trained on data
Software 3.0: programming through prompts and natural language instructions
The key idea is that large language models are becoming a new kind of programmable layer. Instead of specifying every instruction directly, developers increasingly describe goals, constraints, and context, and let the model interpret them.
What Changes for Developers:
In this environment, coding skill still matters, but higher-level judgment becomes more important.
Developers are increasingly responsible for:
defining the problem clearly
structuring tasks for AI systems
evaluating outputs
maintaining quality standards
making architectural and product decisions
Karpathy’s broader point is that the value shifts upward: from syntax and implementation toward judgment, taste, and oversight.
Why Agentic Systems Matter More Than Simple Coding Assistants:
The real leap is not just better code generation. It is the move from single-turn assistance to multi-step autonomous workflows.
That includes systems that can:
plan a sequence of actions
inspect files and context
write and revise code
run tests and interpret failures
continue iterating until they reach a usable result
This is what makes agentic engineering more consequential than vibe coding. It is less about convenience and more about changing the structure of software work itself.
The Limitation: Jagged Intelligence
Karpathy also emphasizes that AI systems remain uneven. They can perform extremely well in some areas and fail unexpectedly in others.
His point is that traditional software automates what can be precisely specified, while language models automate what can be verified.
That distinction matters. It explains why AI can be highly effective in coding environments with feedback loops, tests, and measurable outcomes, while still behaving unreliably in ambiguous or open-ended tasks.
What This Means in Practice:
The best AI workflows are not fully autonomous. They are structured around verification.
That means:
strong testing pipelines
human review
clear task boundaries
systems for checking correctness
The more measurable the output, the more useful the model tends to be.
A More Important Human Role, Not a Smaller One:
One of the most important ideas in Karpathy’s talk is that AI does not remove the need for developers. It changes where human value sits.
As models take over more implementation work, developers become more like editors, reviewers, and system designers. The critical skills become:
taste
prioritization
systems thinking
debugging judgment
knowing when the model is wrong
This is a more strategic role, not a less important one.
Watch What Happens After This Ticker Hits #1
What if you knew today’s biggest stock winner BEFORE it exploded?
Right now, most traders are flying blind — working off data that’s 15, 20, even 30 minutes old. In the stock market, that’s not just slow. That’s the difference between catching a 100%+ gain and watching it happen without you.
But a small segment of “in the know” traders have already figured this out.
They’re using a built-in tool inside the Stocks.News app that most people completely overlook — and it’s quietly handing them an unfair advantage every single morning.
Here’s how it works: Every day before the market heats up, Stocks.News fires up a proprietary scanner that tears through LIVE market data — running it through dozens of indicators to cut through the noise and zero in on the highest-probability breakouts of the day. The result? A ranked Top 5 Stocks list, updated in real time as breakout probability shifts.
Not yesterday’s data. Not delayed feeds. Right now. Live. Constantly recalculating.
And the results have been nothing short of jaw-dropping. Over the past week, nearly every stock that hit the #1 spot went on to surge 100%+ within days. The most recent #1 pick? It climbed 119% in under 72 hours.
This isn’t luck. This is the right data, at the right time, pointing you in the right direction.
❤️ The best way to keep our mission going - at zero cost to you - is by checking out this free guide from our sponsor.
Vibe coding got people excited. Agentic engineering is what actually ships. The difference? Discipline: tests, ownership, clear criteria for "done." Turns out AI raises the bar for good development, it doesn't lower it.