Vibe coding and agentic engineering represent two very different approaches to AI-assisted development. As of early 2026, after several months using Claude Code almost daily on real projects, I can tell you the difference matters. Vibe coding delegates code ownership to the AI. Agentic engineering keeps your engineering judgment in the driver’s seat while AI agents handle the tedious work. For professional developers shipping production code, one of these approaches scales to enterprise requirements, and one does not.
The AI-assisted coding conversation has been dominated by vibe coding, but that term doesn’t capture what’s actually useful for professional developers shipping production code. It’s also led to a lot of people thinking that developers aren’t needed anymore.
Who needs a software engineer when the sales chump can vibe up a web app?
At least it works on their machine! ¯\(ツ)/¯
I think it’s worth cutting through the hype to talk about what it actually is, where it works, and where it falls apart. More importantly, I want to talk about a different definition that doesn’t get as much attention which I think is far more relevant for professional developers: agentic engineering.
What is vibe coding?
Vibe coding is a development approach where you have natural language conversations with an AI and let it build an application for you without reviewing the generated code. The term was coined by Andrej Karpathy in a February 2025 tweet that described exactly this experience: you just talk, the AI builds, and you go with the flow.
Andrej Karpathy coining the term 'vibe coding'
And honestly, for certain scenarios, that’s perfectly fine.
Vibe coding works well for low-risk tasks. Think: collecting data on a form, processing inbound support tickets, or building something that doesn’t perform a bunch of autonomous actions without a human in the loop. It’s also fine when security isn’t a critical concern and you’re not shipping production code that someone has to maintain over time.
Here’s where it gets interesting, though. Vibe coding is also great for people who aren’t engineers. Product managers, marketers, and other stakeholders can use natural language to prototype what they’re trying to express, rather than just showing up with a requirements doc. Imagine collaboratively working with a customer to “vibe up” a prototype of a new feature. That’s a powerful way to communicate ideas, express what you envision in a new app, and get alignment before a single line of production code is written.
Where vibe coding falls short
Let’s be clear: vibe coding isn’t a replacement for engineering.
Vibe coding is great for building applications that are prototypes or for a very small audience, but you can get yourself into some pretty sticky situations if you’re going to roll out these solutions to the entire enterprise.
But when you have a solution that’s more mission-critical and has been rolled out to a lot of people within your organization or to your customers, you need to make sure that you think about maintainability and not just shipping an app, throwing it over the wall, and thinking you’re done. What happens when someone asks for a change or a new feature, or a bug is found? Someone needs to go in and make that change and needs to understand how the application works. Sure, an LLM could analyze the codebase, but it will be missing the original context from the original creation process that might not be fully reflected in the code. The LLM, an AI, is not going to be able to look at an application once it grows to be of a certain size and keep all of it within its context window, at least today.
When you need to review code, understand what it does, and ensure it meets enterprise standards, you need to be more than a conversationalist. You need to be an engineer.
If you don’t understand the code, you can’t debug it, extend it, or explain it to your team.
The cost problem
One area that catches people off guard is cost. These AI tools can write code for you, absolutely. But they don’t always optimize for efficiency. Let’s say the AI generates code that fetches Microsoft Graph data in a loop, making individual API calls for each user instead of batching them. Sure, it works. But Microsoft Graph enforces throttling limits, and a naive implementation can blow through those limits within hours of a production deployment.
Or what if you have API call limits or metered billing? Next thing you know, you’ve completely blown through your entire quota by the end of the first week. These tools write code, but they don’t factor in your operational constraints like API throttling, rate limits, or cost-per-call unless you explicitly tell them to.
Those are the types of things software engineers are familiar with and would know what to look for. But someone who hasn’t built applications and just talks through them? It might get picked up by your coding assistant, but it might not… and you won’t find out until it’s too late.
The “just because you can” problem
Let’s take a SharePoint Framework (SPFx) example. Say your customer wants to customize the user experience on their production SharePoint intranet. You ask your favorite coding AI assistant for a solution, and it suggests using an SPFx application customizer to inject JavaScript and CSS onto modern SharePoint pages to completely rearrange the DOM.
Does it work? Yeah, absolutely. Can you achieve what you’re trying to get done? Sure.
But did your AI assistant factor in that Microsoft explicitly does not support DOM manipulation on modern SharePoint pages? There’s nothing stopping you from doing it, but just because you can doesn’t mean you should. An AI-generated solution can technically execute and silently create an unsupported configuration that voids Microsoft support for the entire intranet tenant, potentially affecting 50,000 users.
This is exactly the kind of judgment call that vibe coding can’t make for you.
Agentic engineering is a better path for professional developers
On the flip side, I think AI makes professional developers significantly more efficient, accurate, and productive when used the right way. That’s where agentic engineering comes in.
Agentic engineering is the practice of using AI-powered coding agents as force multipliers under your direction, while you retain full responsibility for architecture, code quality, and engineering judgment.
For developers building SharePoint Framework solutions, Teams apps, or Microsoft 365 Copilot agents, this distinction matters even more. Platform-specific constraints and enterprise requirements make engineering judgment non-negotiable.
Agentic engineering means using tools like Anthropic’s Claude Code, OpenAI’s Codex, or GitHub Copilot to handle the monotonous and tedious parts of your work. The stuff that, if you could take it off your plate, would let you deliver more value to your users.
Think about scenarios like:
- Major refactoring: Your codebase has grown over time, and you’ve got a 500-line file that’s really hard to read. Let an AI agent help break it into smaller, more maintainable modules.
- Writing unit tests: Let’s be honest, nobody’s favorite task. But before you do a big refactoring, you want tests in place. AI agents are great at generating that initial test coverage and adding a bunch of edge cases that you might not imagine.
- Code review assistance: Have an agent review your pull requests for potential bugs, security issues, or inconsistencies before your teammates even look at it.
- Migration and upgrade tasks: Upgrading a framework version or migrating from one library to another involves a lot of repetitive, pattern-based changes. AI agents can handle the bulk of that work while you focus on the tricky edge cases.
These tools don’t replace your judgment. They accelerate your productivity.
Vibe coding vs. agentic engineering
| Vibe Coding | Agentic Engineering | |
|---|---|---|
| Who it’s for | Anyone — engineers and non-engineers alike | Professional developers |
| Code review | You don’t look at the code | You review, understand, and own the code |
| Risk level | Fine for low-risk, small audience, throwaway projects | Built for production, enterprise-grade work |
| Best use cases | Prototyping, demos, quick internal tools | Refactoring, testing, code review, migrations |
| Engineering judgment | Delegated to the AI | Retained by the developer |
| Maintainability | Not a priority | Core requirement |
Learn More About Agentic Engineering
Interested in learning more about agentic engineering?
Check out Simon Willison’s series, Agentic Engineering Patterns, if you’re interested in learning more about this new development paradigm.
It’s a fantastic series he’s building on over time that’s resonated with me. You can learn more about the series in the introduction about the project in this post: Writing about Agentic Engineering Patterns.
How to think about AI agents in your development workflow
I like to think agentic engineering is a lot like a sous chef. You’re the head chef. You decide what’s on the menu, how it’s prepared, and what goes out to the table.
The coding agent is the sous chef: skilled, capable, fast. But, they are working under your direction. A sous chef doesn’t decide the menu. Instead, they execute, suggest, and handle the prep work.
This is how professional developers should think about coding agents. They’re simply multipliers for you, helping you accelerate your work.
Why professional developers need to learn agentic engineering
I strongly believe developers need to learn these tools and start using them to make themselves more productive. We don’t code with Notepad. We don’t use older IDEs without IntelliSense or integrated documentation. We’ve adopted tools that help with documentation, code completion, and compile-time checks. All of these tools make our lives easier and better.
AI-powered development tools are the natural next step in that same evolution. They’re more of the same kinds of tools we’ve always adopted to work smarter.
In fact, I’ll go so far as to say that if you want to remain relevant as a developer, you need to pay attention to these tools, how they work, and how to use them most effectively.
Bottom line
Vibe coding has its place, but it’s not software engineering. If you’re a professional developer shipping production code, agentic engineering is the approach worth investing in. Keep your engineering judgment at the center, let AI handle the tedious work, and you’ll ship better code faster than ever before.
Here’s a practical way to start: pick one tedious task in your current sprint and hand it to a coding agent. Then review every line it produces. That’s agentic engineering in action.
Microsoft MVP, Full-Stack Developer & Chief Course Artisan - Voitanos LLC.
Andrew Connell is a full stack developer who focuses on Microsoft Azure & Microsoft 365. He’s a 22-year recipient of Microsoft’s MVP award and has helped thousands of developers through the various courses he’s authored & taught. Whether it’s an introduction to the entire ecosystem, or a deep dive into a specific software, his resources, tools, and support help web developers become experts in the Microsoft 365 ecosystem, so they can become irreplaceable in their organization.
