For years I was skeptical about AI tools for design and development.

Every time I tried using large language models (LLMs), I walked away disappointed.

I tested tools like AI coding assistants, wireframe generators, and automated design tools. None of them produced results that were actually usable in real projects.

Instead of helping my workflow, they often slowed it down.

But something changed recently.

Today, I find myself designing with Claude more than using Figma.

And that shift has fundamentally changed how I think about product design, prototyping, and collaboration.

---

My Early Experiences With AI Tools

My first attempts at using AI for design and development didn’t go well.

Last year I experimented with several tools:

• AI coding assistants to modify a small game project
• AI product planning tools to outline features
• AI-generated wireframes for product ideas

The results were almost always disappointing.

The AI outputs were:

• incomplete
• incorrect
• unusable without heavy manual editing

In most cases, it was simply faster to do the work myself.

Looking back, the problem was simple:

I was asking AI to do things I was already good at.

And naturally, the results felt worse.

---

What Changed: Enter Claude

After joining Jane Street, my experience with AI tools shifted dramatically.

Suddenly I was working with technologies that were completely new to me, including:

• OCaml
• Bonsai
• unfamiliar internal systems
• complex data workflows

Because I wasn’t an expert in these areas, AI assistance became incredibly valuable.

Claude helped me:

• understand unfamiliar codebases
• experiment with implementation ideas
• prototype features quickly

But the biggest surprise wasn’t coding support.

It was how much Claude changed my design workflow.

---

My New AI-Powered Design Workflow

Instead of creating long product specs, Figma mockups, and design documentation, I now follow a much more direct process.

My workflow looks like this:

1. Write a short description of the problem and proposed solution
2. Open my editor and start the development environment
3. Prompt Claude with that description
4. Build a quick prototype of the feature
5. Iterate rapidly with AI assistance
6. Deploy it to a development environment
7. Let users try the feature
8. Submit the feature proposal

Instead of designing concepts, I’m building real working prototypes.

And that changes everything.

---

Why Prototypes Beat Mockups

Traditionally, product design workflows look like this:

Problem → Spec Document → Figma Mockups → Engineering Implementation

But there’s always a gap between design intent and real behavior.

Prototypes eliminate that gap.

When you build a working feature, you can:

• test the interaction
• evaluate usability
• validate technical feasibility
• gather real user feedback

You’re not guessing how something will behave.

You’re experiencing it directly.

---

A Real Example: Adding LLM Prompting to JSQL

One example that highlights this shift was a prototype I built recently.

The feature added LLM-powered prompting to a JSQL input field.

(JSQL is an internal SQL dialect used across several tools.)

Instead of writing design docs or mockups, I built a working prototype directly.

With Claude’s help, I iterated on the feature repeatedly.

Over several days I:

• refined the submit button
• added keyboard shortcuts
• improved confirmation messages
• adjusted prompt behavior
• improved the UI feedback flow

These kinds of improvements would normally require multiple design and engineering iterations.

Instead, I could experiment freely.

Claude didn’t get frustrated when I changed my mind.

And that meant I could keep refining the experience.

---

Why AI Prototyping Is So Powerful

AI-assisted prototyping unlocks something designers rarely have:

Unlimited iteration.

Normally, making changes requires coordination between designers and engineers.

That introduces friction.

With AI support, I can explore ideas immediately.

This makes experimentation dramatically easier.

Small improvements that might have taken weeks of coordination can now happen in a single afternoon.

---

How My Use of Figma Has Changed

For a long time, Figma was central to my workflow.

I used it for:

• wireframes
• UI mockups
• design proposals
• collaboration with engineers

But over the past two months, something surprising happened.

My reliance on Figma has dropped dramatically.

Instead of creating visual mockups, I often:

• build interactive prototypes directly
• test real behaviors
• iterate on functionality in code

For some projects, I still start with Figma.

But increasingly, I’m skipping it entirely.

Instead, I design directly in the medium of the final product.

---

AI Has Empowered Designers

One of the biggest benefits of this shift is empowerment.

Traditionally, engineers have the ability to turn ideas into working prototypes quickly.

Designers usually can’t.

Designers often have to convince engineers to build prototypes for them.

That creates a bottleneck.

With AI tools like Claude, designers can now:

• test ideas themselves
• explore feasibility
• present working concepts

This makes it much easier for teams to evaluate ideas.

Instead of debating mockups, people can simply use the prototype.

---

The New Challenge: Reviewing AI-Generated Work

There’s one downside to this new workflow.

When you submit a prototype feature, it often looks complete.

Reviewers might assume the design is final and only focus on code quality.

But that’s not the goal.

The prototype is still a design proposal, not the final product.

To solve this, we treat these prototypes as:

• living design documents
• experimental artifacts
• starting points for discussion

The production implementation often happens later, with engineers owning the final version.

---

The Creative Risk of AI Design

There’s also a creative concern with this workflow.

Designing with AI can push you toward incremental iteration rather than big conceptual leaps.

Instead of imagining entirely new approaches, you may focus on what the AI can easily generate.

This is similar to an older debate in the design world.

Back in the early 2010s, many designers argued about whether designers should learn to code.

Critics believed coding could limit creativity.

Supporters believed it enabled experimentation.

In many ways, AI is bringing that same conversation back.

---

From Pixels to Prototypes

Over time, frontend development became more complex.

Frameworks like React made the development stack heavier.

Many designers—including myself—moved away from coding and focused entirely on tools like Figma.

But now AI is changing that dynamic.

With tools like Claude, designers can again work directly in the medium of software.

Instead of just designing visuals, we can design interactive behavior.

---

The Future of AI-Assisted Design

AI is not replacing designers.

But it is transforming how design happens.

The future of design workflows will likely include:

• AI-assisted prototyping
• design-to-code workflows
• AI-generated UI components
• conversational design iteration

Designers who embrace these tools will be able to move much faster from idea to product.

---

Final Thoughts

Designing with Claude has changed how I work.

Instead of spending days creating mockups and documentation, I can now build working prototypes and test ideas immediately.

It’s not perfect.

And it’s still evolving.

But for the first time in years, I feel like I’m designing in the real medium again.

Instead of describing ideas…

I can simply build them and see what happens.

And that freedom is incredibly exciting.