Choosing your build surface

You have already met most of these surfaces in passing. Canvas inside Gemini, Artifacts inside Claude, Cowork on the desktop, Claude Code in your terminal. There are a few you have not seen yet (Lovable, v0, Bolt) that fill the gap between “ephemeral demo” and “real project”. Knowing where each one lands saves you the cost of starting over.

A spectrum from throwaway to permanent

Pick the surface that matches how long you expect the result to live and how much you want to own it.

• Canvas / Artifacts are the cheapest. The result lives inside a chat. You iterate by talking to it. You demo it once. Maybe you keep the link for a week. The build cost is two minutes; throwing it away costs nothing.
• Lovable, v0, Bolt are the middle. AI-driven app builders that host the result on a public URL. Good for personal small tools and prototypes someone else can click. The platform owns the runtime; you own the prompts and the look.
• Claude Code is the deepest. A real folder on your laptop, real version control, real ownership. Slower start, much bigger ceiling. The right pick when “this needs to live, change, and ship” is the actual goal.

The fourth surface that sometimes fits is Cowork in the Claude desktop app, when the deliverable is a knowledge-work artifact (a working spreadsheet, a deck, a sorted folder) rather than code. Cowork sits between Artifacts and Claude Code in feel: agentic, but pointed at files instead of repos.

When to reach for each

Match the surface to the verb you would use to describe the goal.

• “Show me what this would look like.” → Canvas / Artifacts. You will look at it, react, maybe iterate twice. The result lives in the chat. Charts of CSV anomalies, dashboards for tomorrow’s meeting, clickable mockups, one-shot games to feel out a format.
• “I want a small app I can share a link to.” → Lovable / v0 / Bolt. Small live tool, hosted by the platform. Personal trackers, internal helpers, prototypes you give to a teammate to click. Maintenance is on the platform; you keep iterating on the prompt.
• “I am going to build this and ship it, then change it next month.” → Claude Code. Real folder, real Git, real deploy target. Anything that touches your existing codebase, integrates with your toolchain, or needs to outlive its first week.
• “I need a real file to send” (an .xlsx, a .pptx, a sorted folder of files) → Cowork. Knowledge work where the deliverable is a document or a folder, not a web app.

The trap in both directions:

• Reaching too high. Starting in Claude Code for “I just want to see the chart” is a tax on yourself. Canvas costs you two minutes of typing and you have what you needed.
• Reaching too low. Building a tool you actually want to use in Lovable or Artifacts and then realising you cannot version it, cannot extend it, and cannot move it to your own infrastructure. The migration to Claude Code is real work.

A quick decision matrix

The same question, asked four ways, points at the right surface.

Question	If yes
Will I look at this once and forget it?	Canvas or Artifacts
Do I want a clickable URL I can share with one person?	Lovable, v0, or Bolt
Will this live, change, and ship to real users?	Claude Code
Is the deliverable a document or a folder of files, not a web app?	Cowork

If two answers are yes at once, the lower-on-the-list one wins. “I want to share it AND I plan to ship it” is a Claude Code project, not a Lovable one. “I want to see it AND I want to share the link” is a Lovable thing, not a Canvas thing.

Hands-on

Reflect

• The last three things you tried to build with AI: which surface did you pick for each, and would you pick the same surface again knowing what you know now?
• Which of the four surfaces have you never tried? Pick one and put a 30-minute experiment on the calendar this week.