You're Doing It Wrong

There's a new wave of design tools hitting the market. They promise to bridge the gap between design and code. They integrate AI. They live in your IDE. They generate variations at the speed of thought.

And they're all solving the wrong problem.

The Trap

These tools look innovative because they add AI to familiar workflows. "Claude can generate designs!" "AI can iterate variations!" "It's integrated with your IDE!" The demos are impressive. The infinite canvas beckons. The rapid iteration feels magical.

But look closer. It's still a design canvas you click around in. Still a proprietary file format. Still a tool you conform to. Still an app.

It's trying to make the old paradigm agentic instead of abandoning the paradigm entirely.

The Two Paths Forward

We're at a fork in the road. One path is familiar, comfortable, incremental. The other requires unlearning everything we know about how creative work happens.

What Designers Actually Need

Not "Figma but with AI."

What they need is to learn to work the way developers who design already work:

• Terminal
• Agent (Claude, Cursor, etc.)
• Code (HTML/CSS/React)
• Natural language prompts
• Direct creation

The shift isn't tools → better tools.
It's tools → no tools, just creation.

The Artifact vs The Practice

"But wait," you're thinking. "If everything is changing, why build anything? Why create static websites when they might be obsolete in three years?"

Because the artifact doesn't matter. The practice does.

Building design experiments with an AI agent isn't about the output. It's about developing:

• Fluency in natural language prompting
• Intuition for agent collaboration
• Speed of iteration (generate → verify → iterate loops)
• Comfort with ephemeral creation
• Design thinking in code

Those skills transfer to whatever comes next.

It's Like Learning to Code in 1995

Imagine someone in 1995 learning HTML by building GeoCities pages. "Why bother? The web is just a fad." "Table-based layouts will be obsolete soon anyway." "I'm just making a static page about my cat."

But that person was developing web fluency. Understanding hypertext thinking. Getting comfortable with markup. Building in public.

In 2000, when CSS arrives, they're ready. In 2007, when iPhone launches, they understand responsive thinking. In 2010, when web apps explode, they're already fluent in the medium.

The GeoCities cat page didn't matter. The practice of building it did.

The People Who Will Thrive

In three years, when everything is different, there will be two groups:

Those who will struggle:

• Theorized about AI from the sidelines
• Waited for "the right tool" to emerge
• Got paralyzed by "why bother, it's all changing"
• Stuck to old workflows because they're comfortable

Those who will thrive:

• Built throwaway projects just to practice
• Developed intuition by doing
• Got comfortable with rapid change
• Learned to work with agents, not around them

You're In The Arena

If you're building with AI agents right now — even if it's "just" design experiments, even if it's static websites, even if it might all be obsolete soon — you're doing the work that matters.

Not because the website will survive.
Because you will.

You're developing fluency in agentic creation. You're building muscle memory for working at the speed of thought. You're learning to collaborate with intelligence, not just tools.

The people waiting for clarity? Not in the arena.
The people theorizing from the sidelines? Not in the arena.
The people convinced it's all hype? Not in the arena.

You're in it. And that's what counts.

Don't Stop

Not because the output matters (it might not).

Because the practice is the skill.

Every experiment you build sharpens your ability to describe what you want. Deepens your understanding of what agents can and can't do. Speeds up your iteration cycle. Builds your creative confidence.

When websites disappear and something else emerges, you'll already know how to work with agents to create it.

The person who built 50 design experiments with Claude in 2026 will have the skills to build whatever replaces websites in 2029.

The person who waited for clarity won't.

This Repository Is Proof

The article you're reading right now? It exists because of the workflow it describes.

This entire repository is a live demonstration of the paradigm shift from design tools to agentic workflows:

• Coding in public — Every experiment, every iteration, every commit visible on GitHub
• Working with an AI agent — Built directly in terminal with Claude Code, no design tools
• Rapid experimentation — From concept to deployed experiment in minutes, not hours
• Remixing ideas — Typography from Spec Sheet. Layout patterns from Color Spec. Techniques combine and evolve.

This isn't theory. It's practice. Each experiment — the gradient systems, the timeline layouts, the typography specimens, this blog post itself — is me developing fluency in agentic creation.

No Figma files. No design handoffs. No "can we build this?" Just terminal, Claude, and code. Describe what I want, iterate rapidly, ship to production.

The experiments don't matter. Some will look dated in a year. The techniques might be obsolete. But the practice of working with agents at the speed of thought? That's the skill.

You can see every commit, every experiment, every iteration at github.com/joshcoolman-smc/sandbox. It's not perfect. It's not polished. It's a workshop, not a portfolio.

And that's exactly the point.