Create with Claude: How Sonnet 4.5’s “Imagine with Claude” Signals a New Age of On-the-Fly Building

In a tweet from October 2025, developers were buzzing over “Imagine with Claude”, a new experiment from Anthropic’s Sonnet 4.5. It showed something that used to sound futuristic: describe what you want, the AI scaffolds the code, runs it, lets you interact, and you can tweak it live — all within seconds.

That’s not just impressive. It’s a sign that the traditional wall between building and running software is dissolving. For creators, this means fewer steps, fewer bugs, and an entirely new rhythm of creation.

In this post, I’ll walk through:

• what “on-the-fly code generation” really means
• why current “vibe coding” tools are both exciting and limited
• how “Create with Claude” pushes toward unified environments
• how Davia takes that idea even further — turning the entire workspace into a living, intelligent knowledge system

What “On-the-Fly Code Generation” Really Means

Imagine telling someone: “Build me a small app that shows weather based on a ZIP code, and updates automatically when I change the ZIP.”

In the traditional model, you’d:

• Write code in your editor
• Spin up a dev environment or server
• Test locally
• Push, deploy

On-the-fly code generation compresses that cycle. You prompt the AI, the AI writes code, immediately scaffolds the UI + backend behavior, lets you test and tweak, often right in the same place.

It’s like going to a design studio where you not only sketch, but the canvas becomes live code instantly, and you can “preview” in seconds. No jumping out of your editor, no “build → run → debug → build again” loop.

Why LLMs Already Do a Great Job at Writing Code

Non-technical explanation: Large Language Models (LLMs) are trained on massive amounts of public source code—everything from GitHub projects, npm packages, open datasets.

Because of that, they recognize patterns: how to create login screens, forms, charts, toggles, fetch data from APIs. When asked, they can generate boilerplate or even complex logic by recombining what they’ve already “seen.”

You don’t need to know every detail of how Redux stores work or how Stripe’s API is called—an LLM can suggest code or scaffolding that more experienced developers can refine. For many interactive documents or micro-apps, the scaffolding plus “good enough defaults” is already more than 80% of the work.

The Rise of “Vibe Coding” Tools — And Their Usual Limitations

Over the past few months, platforms like Lovable, Bolt.new, and V0 (by Vercel) have exploded in popularity. They share one goal: make app creation feel effortless. Type what you want, and in seconds you get a live, editable prototype.

But most of these tools still rely on sandboxed environments — isolated spaces where generated code runs safely. This approach works great for heavier use cases like landing pages or production-ready web apps, where full deployment and infrastructure matter.

However, when your goal is simply to build small, interactive documents or micro-apps, that same sandbox setup becomes overkill. Each app lives in its own isolated bubble, making it hard to reuse or connect data between them.

So while “vibe coding” tools have made AI-powered creation intuitive, the current generation still separates the act of building from the environment where that creation lives. That’s what new systems like Claude 4.5 and Davia are starting to solve.

What “Create with Claude” / Sonnet 4.5 Brings to the Table

Anthropic’s Imagine with Claude demo shows the next step: instead of spinning up external sandboxes, the AI builds and interacts within one consistent environment.
You can watch code appear, test it immediately, and keep iterating—all without switching contexts.

This tighter loop unlocks something subtle but powerful: co-creation that feels native.
The agent isn’t just suggesting snippets—it’s evolving a live artifact with you.

It’s a glimpse of a world where “software” is not a static product but a living medium for ideas.

Why Davia’s Unified Build + Runtime Environment Matters

Davia takes that same philosophy but grounds it in a workspace designed for creators.

Templates you can plug & play into your workspace

When someone publishes a template, you can import it—and it becomes part of your workspace instantly. Code, components, and data all come with it.

Agent, code, and data live together

The agent that helps you build and the components it generates share the same context. It can read your data, understand your structure, and suggest changes intelligently.

No friction for small interactive documents

Unlike sandbox-based platforms, Davia doesn’t require separate environments or deployments for lightweight apps. You can build dashboards, forms, calculators, or visualizations that stay live and editable inside your workspace.

A true knowledge system

Every piece of your workspace—content, code, and data—remains connected. That means the agent can understand not only what exists, but how it all relates. It’s less a tool for “building apps,” and more a system for thinking and creating interactively.

The Bigger Picture

AI is collapsing the distance between idea and execution.
As models like Claude 4.5 grow more capable, and as unified environments like Davia mature, creators won’t just “generate apps”—they’ll compose ideas that live, run, and evolve.

• Speed becomes a creative force.
• Templates and prompts become the new code.
• Workspaces become collaborative living systems.

The future of creation isn’t about writing perfect code. It’s about expressing intent—and watching that intent take form in real time.