AI Trends

Vibecoding vs Agentic Engineering

Remy Gieling

February 28, 2026

min read

AI now writes the vast majority of the code at Big Tech — the software factory has been built and Europe hasn't figured it out yet.

In early 2025, Andrej Karpathy — co-founder of OpenAI and former head of AI at Tesla — posted a tweet that rocked the tech world. He described a new way of programming that he vibecoding mentioned: fully embracing the vibes of AI, not even viewing the code anymore, and simply accepting what the model generates. A year later, in February 2026, he looked back on that moment and introduced a new term: agentic engineering. In doing so, he marked not just a name change, but a fundamental shift in how software is created.

For business leaders, this distinction is crucial. Both developments affect your organization — but in very different ways and with very different implications. In this article, I'll explain exactly what vibecoding and agentic engineering mean, why the difference matters, and how you, as a leader, can use both.

What is vibecoding?

Vibecoding is the democratization of software. You describe what you want in plain Dutch (or English), and an AI model generates working code. No programming knowledge required. No IDE. No terminal. Just an idea and a prompt.

With tools such as Lovable, Google AI Studio, Bolt, Replit Agent and Claude, non-technical professionals can now build simple websites, dashboards, internal tools and even small web shops themselves. The threshold for making software has been lowered dramatically in one fell swoop.

Karpathy herself described it as: going fully with the flow, embracing exponential improvements, and forgetting that the code even exists. It is prompt, accept, run, see if it works. Doesn't it work? Paste in the error message and try again. Man is a 'prompt DJ', not an engineer.

Where vibecoding excels

Prototyping and proof-of-concepts: from idea to working product within hours
Personal automation: an HR manager builds an onboarding tracker himself
Internal tools: a marketer creates a dashboard without an IT ticket
Learn and experiment: employees discover technology by doing

The restrictions

Vibecoding also has serious caveats. The generated code is rarely assessed for quality, safety, or maintainability. In practice, this leads to vulnerabilities, technical debt and systems that are difficult to scale up. For a weekend prototype, that's fine. For business-critical systems, it's a recipe for problems.

What is agentic engineering?

Agentic engineering is the professionalization of AI-driven software development. Where vibecoding is about intuition and speed, agentic engineering is about architecture, direction and quality assurance.

Karpathy explains it like this: 'agentic' because 99% of the time, as a developer, you no longer write the code yourself, but orchestrate AI agents that do it. 'Engineering' because it involves art, science and expertise. It's something you can get better at, with its own depth.

Addy Osmani, software engineer at Google Cloud AI and one of the most respected voices in the business, further developed the concept in an influential blog post. He describes the difference succinctly: vibecoding is YOLO. Agentic engineering means that AI does the implementation, but humans remain the owner of the architecture, quality and correctness.

The work process of agentic engineering

Start with a plan. Before you send a single prompt, write a design document or specification. You determine the architecture and break down the work into well-defined tasks. This is exactly the step that vibecoding skips — and where projects go off the rails.
Control, assess results. You give the AI agent a clearly defined task. The agent generates code. You review that code with the same rigor as a pull request from a colleague. Can't you explain what a module does? Then it won't go into production.
Test relentlessly. The biggest difference between agentic engineering and vibecoding is testing. With a solid test suite, an AI agent can iterate in a loop until all tests are successful. Without tests, the agent happily shouts “done” while the code is broken.
Remain the owner. Documentation, version control, CI/CD, monitoring in production — that's still human work. AI accelerates execution, but responsibility for the system isn't shifting.

Why this is relevant to your organization now

In the United States, this shift is already underway. The most advanced engineering teams are already working with multi-agent workflows: they control multiple AI agents in parallel, each specializing in a different task, while engineers are in charge as orchestrators. The question is not whether this will come to Europe, but how quickly.

In the Netherlands and Europe, we are lagging behind in this area. Many organizations are still taking the first steps in generative AI, while the world is already moving towards agentic AI. This is a risk, but also an opportunity — provided you take the right steps now.

“The software developer profession is being fundamentally reformed. The contributions that a programmer makes directly are getting smaller. There's a new programmable level of abstraction that you need to master.” — Andrej Karpathy, February 2026

What does this mean for business leaders?

The distinction between vibecoding and agentic engineering translates into two parallel tracks that you, as a leader, can — and should — activate:

Track 1: Vibecoding for the entire organization

Vibecoding is the way to broad AI literacy. By empowering employees to build simple tools and automations themselves, you're creating a culture of experimentation and ownership. This reduces the pressure on IT, accelerates business innovation, and makes your organization more agile.

Provide Claude, Lovable, or Google AI Studio training for employees to experiment
Organize build days where teams automate their own work processes with Parabola, N8N, or Copilot Studio
Set clear frameworks: vibecoded tools should not work with customer data or business-critical systems, but support a repetitive process
Celebrate successes and share internally how colleagues are making their work smarter with AI

Track 2: Agentic Engineering for IT and Development

Agentic engineering is the future of your IT department. It requires a fundamental redesign of how your engineers work: typing less code, more specifying, orchestrating, and reviewing. This isn't a threat to your developers — it's an opportunity to multiply their impact.

Invest in training: your engineers must learn to work with AI agents, not against them
Work with specs: Get teams started with design docs and specs before starting a single agent
Make testing non-negotiable: A solid test suite is key to turning unreliable agents into a reliable system
Start pilot projects: let a team experiment with tools like Claude Code, Cursor, or GitHub Copilot in an agentic workflow
Redefine the role of your engineers: from writing code to orchestrating systems that write code

The hybrid future

It's not about choosing between vibecoding and agentic engineering. The most successful organizations will use both. Vibecoding for speed and broad adoption, agentic engineering for quality and scalability. The trick lies in knowing when which approach fits.

Industry analysts expect the standard workflow in 2026-2027 to be a human-prompted → agent-executed → agent-reviewed → human approved pipeline becomes: people define the goal, AI agents execute, and people assess and approve. This is not in the future — this is already a daily reality at the most progressive organizations in the US.

Conclusion: The time to act is now

The shift from generative to agentic AI isn't something coming — it's already here. In America, agentic engineering is common ground among the most advanced tech companies. In Europe, we are at a tipping point.

As a business leader, you now have the opportunity to initiate two parallel movements: vibecoding as a democratization tool for the entire organization, and agentic engineering as a new way of working for your IT teams. Those who take both tracks seriously are not only building faster and smarter — but also more future-proof.

“We are seeing the professionalization of AI-driven software development. Where vibecoding caught the excitement of early generative tools, agentic engineering represents a more realistic approach for companies.” — Glide Blog, February 2026