Webflow, Framer and Medusa put agents inside the platform

Webflow, Framer and Medusa each put an agent feature into their own product between 6 August and 1 September 2026, and on 2 September Webflow announced a separate platform built on the same premise. The releases answer the same pressure: teams have spent the year weighing a move off visual builders toward code and coding agents, and we have written about how that move goes when an engineering team runs it. The vendors’ reply is to put the agent where the work already happens.

Webflow: Skills inside ChatGPT and Codex

On 24 August Webflow announced availability in Codex, along with built-in Skills for both ChatGPT and Codex. A Skill bundles the tools, context and practices for a common Webflow job, so a person describes the outcome and the assistant selects the route. Webflow’s own examples are site audits covering SEO and AEO opportunities, accessibility issues, broken links and performance bottlenecks; CMS work such as creating collections and updating items; reviewing custom code before publish; and developer tasks like scaffolding projects, generating reusable components and deploying applications to Webflow. The Skills reach the real site because they call tools in Webflow’s MCP layer, which the company shipped as MCP 2.0 earlier in the summer. Webflow says the ChatGPT integration currently carries no extra cost and does not consume AI credits.

The friction this removes is real. Audits, bulk CMS edits and pre-publish checks are the tasks that quietly consume a marketing team’s week, and describing them in a sentence beats clicking through them.

What it leaves untouched is the shape of the platform. Content stays in Webflow’s database. Editor seats are still priced per person. Collection and item caps are unchanged, hosting stays with Webflow, and the design system still does not leave with you. Those are the reasons on the Webflow migration side of the ledger, and nothing in this window moved them.

Webflow: Source

On 2 September Webflow announced Source, described as “a new platform for modern marketing teams and their AI agents to harness the web”. Webflow’s announcement states that Source makes code the foundation for everything that ships on a site, and that this direct access lets its agents take on larger, more complex changes without the constraints of an abstraction layer. Developers connect an existing codebase and stack to Source, with no migration or rebuild required, according to Webflow.

The governance layer is the part built for teams past the solo-builder stage. The announcement describes each workspace as governed with brand systems, role-based access, approval paths, and audit trails from day one, organised around a “Views” primitive that tailors a workspace to a marketer’s, designer’s, or developer’s specific job. On Webflow’s description, that addresses the collaboration and permissions gap that an assembled stack of an AI coding tool, a headless CMS, and a deployment layer can leave open.

Webflow’s own FAQ keeps Source separate from the platform its customers run today. It says “the platform you know and love isn’t going anywhere,” tells existing customers to keep building on Webflow as they have been, and describes a supported path to Source later, if and when a customer chooses to migrate. Webflow says Source is currently a limited research preview, with select agency partners and customers enrolling in a phased approach over the coming months.

Framer: 3D transforms on the canvas agent

Framer’s update log added 3D Transform support to its Agent on 1 September. The entry lists Rotate XYZ, Perspective, Origin, Depth and Backface among the supported properties, and notes that all 3D layers stay selectable and customizable on the canvas.

The second half is the useful part. Generated design output is worth little when it arrives as something a designer cannot open. Keeping agent-produced layers native to the canvas keeps the designer in charge of the result, which is what a brand review requires.

It is a canvas capability, and it stays inside the canvas. Publishing, hosting, the CMS model and the pricing structure that shape a Framer migration decision are where they were.

Medusa: store data imported from one prompt

On 6 August Medusa published a migration tool in its MCP server that lets a developer prompt an agent to move store data from Shopify into Medusa. By default it covers products, variants, prices and images; inventory levels; and collections and categories. There are two routes. The API-based import pulls from Shopify’s API and is idempotent and resumable, so it can run on a schedule, from an admin page, or via a CLI command. The CSV route reshapes a Shopify product export into the format Medusa’s built-in product importer accepts. The 31 August newsletter, which recapped the v2.19.0 release published 13 August, headlines the tool again and describes a WooCommerce path that moves catalogue, inventory and categories. Medusa shipped v2.20.0 on 2 September and v2.20.1 on 3 September.

This is the most consequential of the three. The most common reason a store stays put is the estimated effort of moving its data, and an import that is resumable and safe to re-run can be rehearsed as many times as the team needs before anyone commits to a date.

The scope of what it delivers is worth being precise about. A data import is scaffolding. The storefront, checkout behaviour, custom apps, payment and fulfilment integrations, and the review and CI layer around all of it remain engineering work, and that work is most of what a replatform costs. Our Medusa development practice starts from that, and the Medusa and Shopify comparison sets out where each platform is the better answer.

What it changes for a decision

The framework holds: who owns the data model, what the platform bills as the team, the catalogue and the content grow, what survives if you leave, and what the site or store has to do that the platform does not offer. Those four questions still stand. Source is the first announcement in this window to argue the underlying point directly: code itself as the artifact a team works on. Webflow’s own FAQ frames Source as a separate platform a customer would migrate to later, and enrolment is still a phased research preview, so none of the four are answered yet.

What did change is the quality of the options on both sides. A team staying on Webflow or Framer now gets meaningful leverage for the price of connecting an integration, and a team that has already decided to leave Shopify has a shorter first mile. The case for staying is stronger than it was in July, mostly for the teams that were staying anyway.

wayf.ai is built and operated this way, with people deciding what changes and agents writing the code inside written conventions. Nothing reaches production until a person has read the diff and looked at the rendered page. We can set the same pipeline up around a client’s site or store, on a platform they keep or one they move to. A free consultation and a scoped, fixed quote come before any commitment.

FAQ

1. Does Webflow's ChatGPT and Codex integration remove the reason to move to code?

   It makes the day-to-day work faster; the platform's structure stays as it was. The Skills make audits, CMS edits and pre-publish checks faster to direct, and for a team staying on Webflow that is a real gain worth switching on. The reasons teams leave sit a layer below: content lives in Webflow's database, editor seats are priced per person, CMS collections and items are capped, and the design system does not travel. Those are the same after 24 August as before it.

2. Can an AI agent migrate a Shopify store to Medusa?

   It can move the data. Medusa's MCP server migration tool covers products, variants, prices and images, inventory levels, and collections and categories, either through an API import that is idempotent and resumable or through a CSV path built on Medusa's product importer. What the prompt does not produce is the store itself. Storefront, checkout logic, custom apps, third-party integrations and the review and CI layer around them are engineering work, and they are where most of a replatform budget goes.

3. Is the output of Framer's 3D agent editable afterwards?

   Framer says it is. The 1 September update states that all 3D layers remain selectable and customizable on the canvas, so rotation, perspective, depth and backface settings the agent applies can be adjusted by hand like any other layer. That matters more than the effects themselves, because generated design work is only useful when a designer can keep directing it.

4. Should we delay a platform decision until these features mature?

   Only if the decision was about tooling speed. The questions that decide a platform are who owns the data model, what the bill looks like as the team and the catalogue grow, what survives if you leave, and what the site or store has to do that the platform does not offer. Webflow's Source announcement on 2 September argues the code-as-artifact point directly, and Webflow says Source is currently in a limited research preview with phased enrolment, so those four answers are where they were across the whole window. If the honest blocker was that day-to-day changes were slow, the new features are worth a fortnight of trying before anyone scopes a move.

5. Does Webflow's Source remove the reason to move to code?

   Not yet, and not for every team. Source is a limited research preview, enrolling select agency partners and customers in phases over the coming months, so judge it once a team can run it. It does change the argument. In Source, Webflow makes code the artifact its agents work on, which is the same premise a move to code was based on. The ownership questions — who holds the data model, what the bill looks like as the team grows, what survives if you leave — are unchanged by an announcement.

Sources

• Webflow, Webflow is now available in Codex and ChatGPT (24 August 2026)
• Webflow, Introducing Source by Webflow (2 September 2026)
• Framer, Updates — Agents: 3D Transforms (1 September 2026)
• Medusa, Migrate your Shopify store to Medusa (6 August 2026)
• Medusa, August 2026 newsletter — v2.19 and the migration tool (31 August 2026)
• Medusa, v2.20.0 release notes (2 September 2026)
• Medusa, v2.20.1 release notes (3 September 2026)
• WAYF, Replacing Webflow with Claude Code: what actually works (1 September 2026)