The Rise of the Agentic Workflow

Matan Dessaur
November 12, 2025 · 6 min read · Web development · AI
Building With Intelligence That Builds Itself
There’s something happening quietly in the background of how we build today.
For years we’ve talked about automation as the big shift. The promise of working faster, cutting repetition, and letting machines handle the boring stuff. But what’s happening now feels different. It’s not just about making things run automatically anymore. It’s about building systems that can actually think with you. Systems that hold context, make connections, and understand what you’re trying to do.
That’s what I mean by an agentic workflow. It’s the next step after automation. Instead of telling a system what to do, you work with intelligent agents that help you figure out how to do it better. They don’t just execute. They collaborate.
The Orchestrated Mind
When I started building my own agentic workflow, I wanted it to feel less like managing software and more like leading a small team. At the center sits my orchestrator agent, the one that keeps everything in sync. It knows who’s doing what and makes sure each step connects to the next.
Around it are five agents that each play a specific role:
- Initializer – sets up the project and the boilerplate.
- Stylist – handles branding and design direction after a short chat about tone and colors.
- Builder – generates components that are functional, clean, and easy to modify.
- Publisher – fills in the content, making sure everything is SEO-friendly.
- Orchestrator – the one that keeps them all working together.
They each live in their own space. Claude Code handles the reasoning and structure. GPT helps me shape ideas and language. Gemini CLI connects everything so that one agent’s output becomes another agent’s input. It’s not perfect, but it feels alive in a way code editors and build tools never have.
When Code Becomes Conversation
The first time I saw the full system working, it didn’t feel like writing code. It felt like a conversation.
I’d tell the orchestrator what I wanted: “Let’s create a site for a local eco-brand. Make it light, breathable, natural.” The Stylist would set the theme, the Builder would create components, and the Publisher would fill the pages with draft content.
What used to take me days now happens in a few hours. Not because the code writes itself, but because the ideas flow faster. I’m not switching between windows or hunting through docs. I’m just thinking out loud and watching the system respond.
That’s the difference. It’s not automation anymore. It’s collaboration.
Ten Days, Ten Sites
To see how far I could push it, I gave myself a challenge: build ten completely different websites in ten days. Each one from scratch. Each under eight hours.
Here’s what came out of it:
siteexample01.netlify.app
siteexample02.netlify.app
siteexample03.netlify.app
siteexample04.netlify.app
siteexample05.netlify.app
siteexample06.netlify.app
siteexample07.netlify.app
siteexample08.netlify.app
siteexample09.netlify.app
siteexample10.netlify.app
Each project felt like a new rehearsal with the same orchestra. Every time, the agents worked a little better together. The Stylist remembered my preferences. The Publisher learned how I write headlines. The Orchestrator got better at predicting what needed to happen next.
By the end, it wasn’t just the system improving. I was learning too. I was learning how to teach it. How to shape instructions that weren’t just technical, but creative.
Teaching Intelligence to Build
Every agent I use starts with a document called agents.md. It’s like a personality file, or a creative brief for the AI. It defines its purpose, its tone, and how it should behave.
It might look like this:
Writing these files feels strange at first. You’re not coding. You’re defining values. You’re teaching the system what matters to you — the things you normally carry in your head when you work. Over time, each agents.md becomes a kind of mirror of your own style. It’s where your rules, instincts, and taste live.
That’s what makes this more than automation. It’s mentorship. You’re training intelligence to build like you do.
The Human Touch
Even with all these systems, I still go over every component by hand. I still adjust the spacing, rewrite the microcopy, fine-tune the animations. That’s the part that keeps me grounded.
The agents handle the repetitive parts. I handle the emotional ones.
It reminds me of a kitchen full of robot sous-chefs. They chop, stir, plate, and clean. But when it’s time to add flavor, balance, and warmth, that’s where the human hand comes in.
The same applies here. Technology gives me speed and precision, but the heart of the project — the craft — still comes from the human behind it.
A New Kind of Collaboration
What I love most about this new workflow isn’t just how fast it is. It’s how collaborative it feels.
Designers can work directly with the Stylist agent and see their color palettes instantly reflected in live components. Marketers can send content briefs to the Publisher agent and get pages back that are optimized and cohesive. Developers can focus on structure and systems instead of rewriting the same layout again and again.
It feels like everyone’s finally working in the same rhythm.
Not siloed, not waiting on handoffs, but connected by a shared layer of intelligence that keeps context for all of us.
The Future That Builds With You
We’re moving into a time where tools don’t just follow instructions. They understand intention. They learn how you work, what you like, and what you care about.
Soon, every creator will have their own small studio of agents trained to their methods, tuned to their voice, and optimized for their projects. It won’t replace people. It will just make every team faster, sharper, and more expressive.
Because intelligence can build almost anything now.
But it still takes a human to build something that feels alive.