How to Actually Use the Framer Agent (Not Just Prompt It)

Most people treat the Framer Agent like a search bar. Type a request, hope for the best, retype it when the result is wrong.

That works occasionally. It is not how the agent is built to be used.

Framer's own team recently laid out six habits that separate a fast, focused agent session from a frustrating one.

We tested each one while rebuilding parts of Frameplate, and the difference is real: the same request produces noticeably better output once you feed the agent the right inputs in the right order.

Here is what actually moves the needle, based on Framer's guidance and our own time in the editor.

1. Give it context before you give it instructions

Select the layers you're working on before you type anything.

The agent reads your selection as scope. Without it, the agent is guessing which part of the page you mean, and it will often edit the wrong section or apply a change too broadly.

This is the single easiest habit to build and the one most people skip.

2. Show, don't just tell

You can upload images, files, or paste a URL to point the agent at a specific look or structure you want replicated.

If you're trying to describe a layout in words, you're doing it the hard way. A reference screenshot or a competitor's URL gets you there faster and with fewer corrections.

3. Use / for scoped actions

Typing / brings up focused commands for layout, CMS work, and styling.

These are narrower than a freeform prompt, which means less room for the agent to misinterpret what you want.

If you know exactly what category of change you're making, use the slash command instead of describing it in prose.

4. Match the model to the job

Framer lets you switch between GPT, Sonnet, Opus, and Fable 5 depending on the task. Heavier reasoning jobs (structural changes, CMS logic) benefit from the stronger models.

Quick styling tweaks don't need that overhead, and using a lighter model there saves both time and credits.

Treat model choice as part of the workflow, not an afterthought.

5. Use @ to point at something specific

Typing @ lets you reference an exact page, CMS collection, style, or asset instead of describing it.

If your instruction depends on something that already exists in your project, mention it directly rather than explaining it. It removes an entire category of ambiguity.

6. One chat, one task

Starting a new chat for each task keeps the agent's context clean.

Reusing a single thread across unrelated tasks is the most common way people end up with an agent that seems to "forget" what it's doing or drift off task.

If you're moving to a new part of the project, start a new chat.

The pattern underneath all six

None of these tips are really about the agent being smarter or dumber.

They're about giving it less to guess. Selection, references, mentions, and fresh chats all do the same job: they narrow the problem down to something the agent can actually solve well, instead of leaving it to infer your intent from a single sentence.

If you've been getting inconsistent results from the Framer Agent, it's worth testing this against your own last few sessions. The fix is rarely a better prompt. It's usually a smaller, better-scoped one.