Over the past couple months, I’ve been diving into two emerging paradigms in the world of AI-assisted development: Agentic Coding and Vibe Coding. Both offer fascinating, albeit very different, approaches to working with intelligent systems. I wanted to document my early experiences to help clarify what each of these methods offers — and how they’ve influenced the way I build.
🧠 What Is Agentic Coding?
Agentic coding is all about giving AI systems goals rather than tasks. You describe what you want, and the agent figures out how to do it — often breaking it down into subtasks, iterating, and self-correcting as it goes. It’s not just autocomplete on steroids — it’s like having a junior, maybe even senior, engineer who can think, plan, and code on your behalf.
💻 My Experience So Far (Agentic)
At first, using agentic systems felt like magic — and mind-blowing. I'd give it a goal like “Add a notification system,” and the system would reason through the steps, generate the code, and even unit test the flow.
I found myself stepping into more of a product owner role, guiding the agent, reviewing its work, and steering its decisions.
Upsides:
Downsides:
That said, when paired with tight version control and a “trust but verify” mindset, it’s incredibly powerful — especially in fast-paced development cycles.
🎨 What Is Vibe Coding?
If agentic coding is about outsourcing the “how,” vibe coding is about expressing the feel of what you want and letting the AI fill in the details. It’s less like writing blueprints and more like painting in broad strokes.
You might say, “I want the galaxy to be a spiral like the Milky Way,” and the system codes a procedural generation system with spiral arms, a galactic center, and a visually accurate structure — and does it fast.
🚀 My Experience So Far (Vibe)
Vibe coding has felt incredibly intuitive — and liberating — especially for creative work.
When I’m building assets or UI for my game Stellar Throne, I don’t always know what the final design should look like. Vibe coding lets me describe the tone, mood, and visual direction, and get results that are often better than I imagined.
Highlights:
Great for prototyping game visuals, UI layouts, procedural content
Fast iteration and creative discovery
Feels more like collaboration than delegation
Challenges:
Sometimes results are too literal
Editing images or visuals requires learning some unique language
But once you get the hang of it, the model usually “gets” you
Vibe coding also pairs beautifully with agentic systems. I can vibe-code a UI mockup, then let an agent implement the layout and wire up the logic. That hybrid workflow? It’s starting to feel like the future.
🧩 Wrapping Up
As someone building a complex project like Stellar Hegemony, I’ve found both Agentic and Vibe coding to be more than just tools — they’re new creative mindsets. They’ve expanded not just how I work, but how I think about the work. Less time buried in syntax, more time focused on vision and design.
These paradigms aren’t magic bullets. They require learning, judgment, and a willingness to trust (but verify). But when they click, the results are transformative.
If you’re curious, try integrating one of them into your workflow. Start small — maybe prompt an agent to build out a feature, or vibe-code a mockup of a screen. See what happens. You might find, like I did, that you’re no longer just coding — you’re collaborating with something new.
And that’s a pretty exciting place to be.
Thanks for reading — and if you’re experimenting with these approaches too, I’d love to hear what’s working (or not) for you.
— MrPhil
