Framer 3.0: AI Agents, Branching, and Community – What Webflow Doesn't Have Yet

On June 16, 2026, Framer unveiled Framer 3.0 during its “Building Agents for Framer” event. The announcement introduced AI Agents that work directly on the canvas, Branching for safe iteration without affecting production, External Agents that can control projects from tools like Claude Code or Cursor, and a complete overhaul of the Community experience.

‍

As a Webflow expert, I looked at these updates with a very specific question in mind: what is genuinely new, and more importantly, what can Framer do today that Webflow still cannot?

Here’s a clear summary, without the hype, along with my perspective as someone who builds websites every day.

The timing is also noteworthy. Framer 3.0 arrives just as Webflow is increasing its pricing and restructuring parts of its organization, a topic I'll come back to later.

‍

Watch Framer 3.0 in Action (Official Demo)

The tutorial below demonstrates what these AI agents can actually do in a production workflow: bulk-replacing images, auditing website copy, converting repeated elements into reusable components, managing mobile navigation, connecting sections to the CMS, and much more.

‍

‍

1. Framer Agents: AI That Edits on the Canvas, Not in a Chat Box

This is the core of the announcement. Framer 3.0 introduces a new Agent tab alongside the traditional Style panel, complete with a model selector (Opus 4.8 in the demo). But the real story isn't "AI that generates content", it's AI that generates native, editable design elements directly on the canvas.

In practice, the agent can:

• Create entire sections and pages that match the rest of the website, using existing styles, structure, and content as context.
• Target a specific frame or layer through a contextual selection tool. Click an element and it becomes part of the prompt. Pages can also be referenced using @.
• Make an entire website responsive in one go by generating tablet and mobile variants of components such as navigation menus and footers.
• Add refined, staggered animations including fades, blurs, character-by-character, word-by-word, and scroll-triggered effects, then refine them through conversation.
• Manage light and dark themes, refactor typography styles, standardize heading scales, and clean up color systems.
• Generate functional code components based on existing canvas elements. For example, a static analog clock can become a live clock with a configurable timezone property.
• Create design-oriented pages such as style guides, sticker sheets, and hero section explorations featuring animated shaders.

The key point for professional workflows is that the output is production-ready. There's no code to inspect and no cleanup required. You generate, review, and publish.

The collaboration model is equally important. You can let the agent generate content, manually rearrange elements, hand control back to the agent, and use undo, redo, and revert controls for every prompt. A Follow mode even lets you track the agent while it works elsewhere in the project.

Framer co-founder Jorn van Dijk summarized the philosophy well: AI can quickly get you to a first version, but refining that version is where the real work happens. Designers compare, move, adjust, and evaluate directly on the canvas, not inside a chat interface. It's difficult to argue with that perspective.

‍

The CMS Agent: The Most Surprising Feature

The second agent received less attention during the keynote but could have the biggest impact on production work.

It can:

• Create a CMS collection from existing page content and dynamically connect sections to it, such as displaying six services from a collection of thirteen.
• Restructure relationships between collections, including multi-reference connections between content types like Work and Team.
• Bulk rename items, transforming rough working titles into polished article titles.
• Simplify slugs and automatically manage redirects from old URLs to new ones.

This is exactly the kind of repetitive work that consumes hours on large projects. Here, it's reduced to a handful of prompts.

‍

2. Branching: Iterate Without Touching Production

Framer 3.0 introduces Branching, arguably the most strategic feature for teams.

You create a branch, effectively a sandbox environment, where major changes can be tested safely. Whether you're making an entire website responsive or experimenting with a completely new design direction, everything happens away from production.

You can compare versions, review changes, and merge them into the main project with a simple Apply to Main action. The live website remains untouched until you're ready.

Combined with AI agents, this solves one of the biggest concerns around AI-powered production workflows: you can allow large-scale automated changes without risking your live website.

Other collaborators can also work within the same branch, making the workflow suitable for teams rather than individual creators.

‍

‍

3. External Agents: Control Framer From Your Terminal

This is the feature that will resonate most with technical users.

Beyond native agents, Framer now supports External Agents. You can connect Claude Code, Cursor, Codex, or virtually any LLM through the Framer CLI using /framer and a project link, allowing you to control projects from outside the editor.

The most impressive demo showcased a chaotic folder containing CSV files, Markdown documents, and scattered images being transformed into a fully relational CMS structure with five collections, reference fields, and correctly mapped assets, all through a single prompt and within its own branch.

Additional capabilities include:

• Parallel audits for performance, SEO, and accessibility, exported as Markdown reports.
• MCP server integrations with tools such as Notion, Figma, and X for importing and exporting content, styles, and design tokens.
• The ability to work across multiple Framer projects simultaneously.

The broader idea is simple: users benefit from the latest advances in AI models immediately, without waiting for Framer to integrate them natively.

‍

4. Community Refresh and Hackathon

Framer is also consolidating its ecosystem through a redesigned Community experience.

The new platform includes:

• A social feed for sharing work.
• Member profiles for discovering creators.
• Community contests.
• An enhanced marketplace with comments, likes, and a simplified publishing process.

The entire interface has been redesigned.

To launch the new ecosystem, Framer announced a 24-hour hackathon with $100,000 in prizes.

‍

‍

5. What About Webflow? What Framer Does That Webflow Still Doesn't

This is the question many people will ask after watching the keynote.

To answer fairly, Webflow has also made significant progress with AI throughout 2026.

‍

What Webflow Already Offers

• AI Assistant (currently in beta), capable of generating and refactoring sections while understanding site context.
• AI Site Builder, which can generate a multi-page website from a prompt (not that great).
• AI-powered SEO and AEO tools, including alt text generation, metadata creation, schema markup, Optimize, and Analyze (this is quite good).
• A Webflow MCP integration for connecting Claude and operating at scale (great for CMS content generation).
• Real-time collaboration, generally available since February 2026.

‍

What Framer 3.0 Adds That Webflow Still Lacks

- Canvas-native site-wide agents

Webflow's AI Assistant remains beta-only, limited to AI Site Builder projects and certain templates. Framer's agents can perform sweeping operations across entire websites, such as global responsiveness, site-wide copy audits, and full theme refactors.

- Branching with merge workflows

Webflow offers staging environments, backups, and version history, but it lacks a true branching, comparison, and merge system directly within the design environment.

In my view, this is the most significant gap and arguably the most important capability for teams looking to adopt AI safely.

- A CMS agent built from scratch

Creating an entire relational CMS structure from folders of files, automatically mapping assets and generating references, goes beyond what Webflow currently offers natively.

- A more integrated External Agents workflow

Webflow has MCP support, which provides partial parity, but the combination of Framer CLI and canvas-level control from the terminal appears more mature based on the current demonstrations.

To be fair, Webflow still maintains strong advantages in areas such as localization, enterprise adoption, and serious e-commerce workflows, particularly when paired with Shopify, a space I know well.

These are increasingly two platforms competing for the same audience.

‍

6. The Webflow Context: Price Increases, Restructuring, and AI Frustrations

Framer 3.0 launches at a particularly interesting moment for its competitor.

In 2026, Webflow increased pricing and later announced a company restructuring. CEO Linda Tong published a statement titled "Evolving Webflow for the Agentic Web", confirming workforce reductions as part of a broader strategic shift.

While exact figures were not officially disclosed, industry estimates suggested approximately 140 positions were affected, around 8% of the company's workforce. The move mirrors a wider trend across the no-code industry, following similar actions by companies such as Wix and ClickUp.

‍

Webflow's repositioning is clear: rather than competing on simple website creation, a space increasingly covered by lightweight AI builders, it aims to become a comprehensive marketing platform for the agentic web.

Beyond the strategy, however, many practitioners share a more practical frustration: Webflow's AI capabilities still feel limited in day-to-day production work.

This is precisely the point made by Timothy Ricks in a recent video, and his analysis aligns closely with my own experience.

‍

‍

Timothy Ricks' Perspective: Why AI in Webflow Still Feels Limited

His arguments will resonate with anyone who works in Webflow every day.

- The Webflow MCP is slow, token-intensive, and tends to generate fairly basic designs. Claude is not operating in its ideal environment when interacting with a proprietary API. It performs far better when writing and reviewing code in real time than when issuing commands such as "add a heading element to the page."

- Once a section exists inside Webflow, AI struggles to keep iterating on it. In many cases, the entire section needs to be reimported. This remains one of the biggest friction points in the workflow.

- Code Components live in their own ecosystem. They are often built with React or Tailwind and use styling systems that differ from the rest of the website. In practice, this can feel similar to placing everything inside a code embed, which goes against one of Webflow's core strengths: maintaining a unified visual development environment.

Many users rely on Code Components simply because they are currently the most practical way for AI to continue modifying an existing section after it has been added to a project.

‍

For personal projects, AI-assisted coding can make sense. Webflow bandwidth can become expensive, and personal projects generally don't require polished collaborative editing experiences. For client work, however, the tradeoffs are far less attractive.

‍

Where Webflow remains exceptionally strong is in its visual builder and design system architecture, particularly features such as variants, properties, and slots.

According to Timothy, no serious AI-native competitor has yet matched Webflow in this area. He continues to monitor emerging tools such as Ship Studio, which recently became open source. Interestingly, spending more time working directly in code only reinforced his appreciation for the value of a visual editor.

‍

My Analysis: This Is Exactly the Gap Framer 3.0 Is Targeting

The biggest frustration Timothy describes, the inability for AI to natively iterate on an existing section without rebuilding or reimporting it, is precisely what Framer Agents are designed to solve.

They work directly on the canvas using native Framer elements, allowing designers to continue collaborating with the AI section by section and page by page.

Where Webflow often requires a combination of MCP workflows, reimports, or Code Components, Framer makes AI-assisted iteration the default experience.

‍

That doesn't mean Webflow should be abandoned.

‍

Webflow's visual development experience and design system capabilities remain outstanding. Timothy's broader advice is particularly relevant during the current AI transition: don't compete on speed and price alone.

AI inevitably pushes the market toward those metrics.

A better strategy is to move upmarket, focusing on projects that require accessibility, robust design systems, scalable architecture, and strategic thinking. That's where freelancers and agencies continue to create real value regardless of which platform they use.

‍

My Perspective as a Webflow Expert

Framer 3.0 doesn't "kill" Webflow.

What it does is put pressure on two areas where Webflow will likely need a response:

1. Branching workflows with comparison and merging
2. AI agents capable of working across an entire production website

Most other capabilities, such as AI-generated sections, AI-powered SEO, and MCP integrations, already exist on both platforms, albeit at different levels of maturity.

For freelancers and agencies, the real question isn't Framer or Webflow?

‍

The real question is:

Can I save time in production and maintenance without sacrificing control?

Based on what Framer demonstrated, the combination of canvas-native agents and branching workflows is a very compelling answer to that question.

I'll be testing these features extensively on a real project and will share a practical, hands-on review once I've spent more time with them.

‍

Going Further

The event replay and full recap: framer.com/events

Framer 3.0 release notes: framer.com/updates

‍