# Webflow vs Framer vs building with AI: costs and trade-offs

> Webflow and Framer are the better choice when designers own the site and nobody will maintain code: they cost less in subscriptions, Framer most of all, and editors change content directly on the page. The coded stack (an AI coding agent such as Claude Code, Codex or Cursor, a headless CMS such as Diggama, Cloudflare hosting) is worth it when you want to own the code, outgrow per-seat, per-locale or bandwidth pricing, and have someone who can review diffs. In our three scenarios, three years of subscriptions cost $360 to $3,600 on Framer, $900 to $3,168 on Webflow and $1,803 to $4,181 on the coded stack with a $200-a-year agent plan, and build and maintenance hours usually outweigh all of them.

Source: https://diggama.com/guides/webflow-vs-framer-vs-ai-coding/
Last updated: 2026-10-07

This page compares three ways to run a marketing site in late 2026: **Webflow**, **Framer**, and what we will call **the coded stack**: a codebase that an AI coding agent writes (Claude Code, OpenAI Codex, Cursor or another), content in a headless CMS (Diggama in our examples), and hosting on Cloudflare. It covers what each costs over three years in three realistic scenarios, with the math shown, and the trade-offs that money does not capture.

Prices are list prices checked on October 7, 2026, in US dollars, before tax. Webflow and Framer are billed yearly. Diggama prices are in euros and converted at the [ECB reference rate](https://www.ecb.europa.eu/stats/policy_and_exchange_rates/euro_reference_exchange_rates/html/index.en.html) of that day, 1 EUR = 1.1177 USD.

## What is the difference between Webflow, Framer and an AI coding agent?

Webflow and Framer are hosted visual builders: you design on a canvas, the platform hosts the site, and editors change content on the page. An AI coding agent such as Claude Code, Codex or Cursor works differently: it writes a normal codebase that you own and host where you like, so you need a separate CMS for editors and someone to run the agent when the layout changes.

| | Webflow | Framer | Coded stack |
|---|---|---|---|
| How you build | Visual canvas close to HTML and CSS | Visual canvas close to a design tool | Prompts to your agent, in a terminal or editor; it writes Astro or Next.js |
| Who edits content | Editors on paid seats, on the page | Content editors on paid seats, on the page | Editors in the CMS dashboard, as forms |
| Hosting | Webflow, included in the site plan | Framer, included in the site plan | Cloudflare Workers, free plan available |
| Code you own | HTML and CSS export, no CMS pages | None, no HTML export | All of it, in your Git repository |
| AI | Webflow AI, MCP server | Framer agents, external agents (preview) | Any coding agent, plus the CMS's MCP server |
| Pricing grows with | Seats, bandwidth, locales | Seats, pages, CMS items, bandwidth, locales | CMS users, content types, API keys |

The last row matters more than the headline prices: the cheapest option depends on how many people edit, how much traffic you have, and how many languages you publish in.

## How much does each option cost over three years?

On subscriptions alone, Framer is the cheapest of the three in two of our three scenarios, Webflow is cheapest for the multilingual site, and the coded stack is the most expensive in all three. The gap ranges from about $1,000 to $2,000 over three years. What usually decides the total is the hours spent building and maintaining the site, which no subscription includes.

### The prices used

| Item | Price (per month) | Source |
|---|---|---|
| Webflow Premium site plan (CMS, 50 GB bandwidth) | $25 billed yearly, $39 monthly | [Webflow pricing](https://webflow.com/pricing) |
| Webflow Limited seat (edit content) | $15 per seat billed yearly | Webflow pricing |
| Webflow Localize Essential | $9 per locale billed yearly, up to 3 | Webflow pricing |
| Framer Basic / Pro | $10 / $30 billed yearly | [Framer pricing](https://www.framer.com/pricing) |
| Framer content editor / locale | $10 per editor / $20 per locale | Framer pricing |
| Claude Pro / Max (include Claude Code) | $17 billed annually ($200 a year) or $20 monthly / from $100 | [Claude pricing](https://claude.com/pricing) |
| ChatGPT Plus / Pro (include Codex) | $20 / from $100 | [Codex pricing](https://learn.chatgpt.com/docs/pricing) |
| Cursor Pro | $20 | [Cursor pricing](https://cursor.com/pricing) |
| Cloudflare Workers Free | $0, static asset requests free and unlimited | [Workers pricing](https://developers.cloudflare.com/workers/platform/pricing/) |
| Diggama Starter / Growth | 29.90 EUR / 89 EUR (about $33.42 / $99.48) | [Diggama pricing](/pricing/) |

Assumptions that apply to every scenario:

- **36 paid months, no price changes.** Free trials and free plans used while building are left out.
- **Domains are excluded.** You pay for one in every case, except that Framer includes a complimentary domain on yearly plans.
- **One AI agent subscription** for the person who maintains the coded site, counted in full at Claude Pro billed annually, the cheapest of the main agents' usual paid plans: $600 over three years. The next section shows the range.
- **Webflow on its free Starter Workspace**, which includes one full seat for the owner. Seats for editors are bought on top. If you want code export, add the Core Workspace at $19 a month billed yearly.
- **Hosting on the Cloudflare Workers Free plan**: 100,000 Worker requests a day and 3,000 build minutes a month ([Builds limits](https://developers.cloudflare.com/workers/ci-cd/builds/limits-and-pricing/)). Prerendered pages are static assets, so only the form endpoint uses Worker requests. Workers Paid starts at $5 a month.

### The AI agent line

The coded stack needs one subscription to a coding agent. The three main options, Claude Code, OpenAI Codex and Cursor, cost about the same on their usual paid plans:

| Agent | Plan | Per month | 3 years |
|---|---|---|---|
| Claude Code | Claude Pro, billed annually | $17 ($200 a year) | $600 |
| OpenAI Codex | ChatGPT Plus | $20 | $720 |
| Cursor | Cursor Pro | $20 | $720 |
| Claude Code or Codex, heavy use | Claude Max or ChatGPT Pro | from $100 | from $3,600 |

The scenarios below count the cheapest row, Claude Pro billed annually. ChatGPT Plus or Cursor Pro adds $120 to each coded total, and a $100 tier kept for three years adds $3,000. OpenAI's pricing page also lists Codex on ChatGPT Free and Go ($8 a month), but the [Codex README](https://github.com/openai/codex) recommends signing in with Plus or a higher plan, so the table starts there. Two secondary options cost less: [GitHub Copilot Pro](https://github.com/features/copilot/plans) is $10 a month and includes agent mode and MCP, and [Gemini CLI](https://github.com/google-gemini/gemini-cli/blob/main/docs/resources/quota-and-pricing.md) has a free tier of 1,000 requests a day with a Google account. They bring the AI line to $360 or $0. At $0, the coded total in scenario 3 drops just below Framer's; the other rankings do not change.

### Scenario 1: a 5-page marketing site with a blog

One person builds the site and writes the posts. Five static pages, a blog, a contact form, modest traffic.

| Line | Webflow | Framer | Coded stack |
|---|---|---|---|
| Site and hosting | Premium, $25 x 36 = $900 | Basic, $10 x 36 = $360 | Cloudflare Workers Free, $0 |
| Seats | Owner's full seat included, $0 | Workspace owner included, $0 | Not applicable |
| CMS | Included | Included (2 collections, 1,000 items) | Diggama Starter, 29.90 EUR x 36 = 1,076.40 EUR = $1,203 |
| AI tool | Webflow AI credits included | Framer AI credits included | AI agent (Claude Pro), $200 x 3 = $600 |
| **3-year total** | **$900** | **$360** | **$1,803** |

Webflow's Basic plan ($15 a month) has no CMS, so a blog needs Premium. Framer Basic allows 30 pages and 2 CMS collections, which is enough here.

For a site this size, the honest answer is that a one-person team does not need a CMS on the coded stack at all: keep the posts as Markdown in the repository and the total drops to $600. A CMS starts paying for itself when someone who does not use Git has to publish. The [guide on why AI-built websites need a CMS](/guides/why-ai-built-websites-need-a-cms/) covers that threshold.

### Scenario 2: a 50-page site with a CMS and 3 editors

A company site: about 50 pages, a blog, team pages, a contact form. One person owns the build, three marketers edit and publish content.

| Line | Webflow | Framer | Coded stack |
|---|---|---|---|
| Site and hosting | Premium, $25 x 36 = $900 | Pro, $30 x 36 = $1,080 | Cloudflare Workers Free, $0 |
| Editor seats | 3 Limited seats, $15 x 3 x 36 = $1,620 | 3 content editors, $10 x 3 x 36 = $1,080 | Included in CMS users |
| CMS | Included (20,000 items, 40 collections) | Included (2,500 items, 10 collections) | Diggama Growth, 89 EUR x 36 = 3,204 EUR = $3,581 |
| AI tool | Included credits | Included credits | AI agent (Claude Pro), $600 |
| **3-year total** | **$2,520** | **$2,160** | **$4,181** |

Four people need dashboard access (the builder and three editors), which is above Diggama Starter's 3 users, so this scenario uses Growth: 6 users, 10 custom blueprints and 3 API keys. Three keys give each consumer its own token: a read token for the build, a create-only token for the form, and one to spare. With two editors, Starter fits and the coded total falls to $1,803. Its single API key then has to serve both the build and the form endpoint, so that one token carries the abilities of both. It works, but a leaked key exposes everything it can do, which is the practical reason to pay for separate keys.

The Webflow column assumes the company owns the Workspace. If a freelancer or agency builds and hosts the site in their own Workspace, Webflow's freelancer and agency Workspace plans include one to three free client seats per site ([Webflow pricing](https://webflow.com/pricing)), which lowers or removes the seat line.

### Scenario 3: the same site in three languages

Scenario 2 published in English plus two more languages, with the same four people.

| Line | Webflow | Framer | Coded stack |
|---|---|---|---|
| Site and hosting | Premium, $900 | Pro, $1,080 | Cloudflare Workers Free, $0 |
| Editor seats | 3 Limited seats, $1,620 | 3 content editors, $1,080 | Included in CMS users |
| Localization | Localize Essential, 2 locales, $18 x 36 = $648 | 2 locales, $20 x 2 x 36 = $1,440 | Diggama variants, no extra fee |
| CMS | Included | Included | Diggama Growth, $3,581 |
| AI tool | Included credits | Included credits | AI agent (Claude Pro), $600 |
| **3-year total** | **$3,168** | **$3,600** | **$4,181** |

Webflow's Essential tier serves locales in subdirectories (`/fr/`) and includes 10,000 AI-translated words per locale per month. Localized URL slugs, per-locale images and automatic visitor routing need Localize Advanced at $29 per locale, which brings Webflow to $4,608. In Diggama, a blueprint declares its language [variants](https://docs.diggama.com/blueprints) and every API call takes a `variant` parameter. Translation is not a separate product: your agent does it through the MCP server, whose "Translate a record" prompt (a slash command in clients that support MCP prompts) copies a record into another variant and translates only its text fields.

### What changes the ranking

The scenarios above favor the builders because they are small. Each platform charges for something different as a site grows:

| When you add | Webflow | Framer | Coded stack |
|---|---|---|---|
| Traffic beyond the included bandwidth | Premium: 100 GB for +$20/mo, 400 GB for +$140/mo | Pro: $40 per extra 100 GB, 2 TB max | No charge for static asset requests |
| Another editor | $15/mo per Limited seat | $10/mo per content editor | Free up to the plan's users (3, 6 or 12) |
| More pages | 300 static pages on Premium | Pro: 150, then $20 per 100, 700 max | No per-page pricing |
| Another language | $9 or $29 per locale per month | $20 per locale per month | No per-language fee |
| More content types | 40 collections on Premium | Pro: 10, then $40 per 10, 40 max | 5, 10 or 25 custom blueprints by plan |

A site with 400 GB of monthly traffic pays Webflow $165 a month for the Premium plan with its 400 GB tier, and Framer $150 for Pro plus three extra 100 GB blocks, against $0 for Cloudflare static hosting. Webflow also upgrades the plan automatically after two consecutive months over the bandwidth limit. That is where the coded stack's fixed costs start to win.

Subscriptions are also the small line next to the hours spent building and maintaining a site. The free [website cost calculator](/tools/website-cost-calculator/) takes your own hourly rates, maintenance hours and a one-off migration, and compares three-year totals.

## Which gives you more design freedom?

The coded stack has no ceiling: anything a browser can render, your agent can write. Webflow and Framer have ceilings, but they are high, and they let a designer reach a polished result by hand, without describing it in words.

**Webflow** exposes the CSS box model, classes and variables almost directly, plus interactions built on GSAP. Its Premium plan adds code components, so developer-built or AI-generated components can be placed directly on the canvas. A designer who knows CSS rarely hits a wall.

**Framer** feels like a design tool. Layout, animation and responsive breakpoints are quick to get right, which suits launch pages and portfolios. It is less suited to deep page structures or large content models: Pro tops out at 40 collections and 40,000 CMS items even with add-ons.

**The coded stack** can do anything: custom search, a pricing calculator, an authenticated area, a content model of any shape. The constraint moves to the person prompting. Every coding agent works best from a clear reference (a screenshot, a Figma frame), and without a designer reviewing the result, coded sites tend to look generic.

## Which is easiest for marketers and editors?

Webflow and Framer are easier for editors: they click on the live page and type. On the coded stack, editors work in the CMS dashboard with forms, and see their change on the site after a rebuild. That is a different habit, and some marketers will miss on-page editing.

What editors get in Diggama, from its documentation:

- **Fields, not pages.** Each content type (a blueprint) is a form: text, rich text, images, slugs, relations. Editors cannot break the layout, because the layout is not theirs to change.
- **Drafts, publishing and scheduling.** Drafts autosave; [publishing](https://docs.diggama.com/publishing) can be immediate or scheduled.
- **Roles.** Editors can create, edit and publish content, but cannot change blueprints, tokens or workflows.
- **An SEO panel**, when enabled on a blueprint, that scores each record as you type, and a shared media library.
- **A Page builder field** for landing pages assembled from a fixed catalogue of blocks, for teams that want some layout control without code.

What they do not get is a WYSIWYG view of the real site. A published edit is not live instantly: a Diggama Workflow calls the Cloudflare deploy hook on every save, and the change appears when that build finishes. Reviewing drafts in context needs a separate preview build, as the [pillar guide](/guides/build-a-website-with-ai/) shows. If your marketers build new landing pages every week without a developer, Webflow or Framer will serve them better.

## Who maintains the site, and what happens when that person leaves?

With Webflow or Framer, the platform maintains hosting, security and the runtime, and any designer who knows the tool can take over the site. With the coded stack, someone has to own a repository: dependencies, framework upgrades, failed builds. An AI agent makes that work faster, but a person still has to read the diffs and decide.

This is the trade-off people underestimate. When the builder leaves, the site keeps running; the next change is the problem. Someone new has to find the repository, the Cloudflare account, the CMS tokens and the reasons behind the code.

Four habits make that handover routine:

1. **Accounts belong to the company.** The Git organization, the Cloudflare account and the Diggama project are owned by a company account, with at least two administrators.
2. **The project explains itself.** An `AGENTS.md` file at the root holds the rules any agent follows in this repository (where content comes from, which commands to run, what never to hardcode), so the next person can use a different agent from the last one. Codex, Cursor and Copilot read it natively. Claude Code [reads it](https://code.claude.com/docs/en/memory#agents-md) only when there is no `CLAUDE.md`; otherwise put `@AGENTS.md` at the top of `CLAUDE.md`. Gemini CLI needs `AGENTS.md` added to [`context.fileName`](https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/gemini-md.md). The [AGENTS.md template for websites](/guides/agents-md-template-for-websites/) is a starting point.
3. **Content is not in the code.** If editors work in the CMS, a change of developer does not touch their workflow at all.
4. **The stack is boring.** Astro or Next.js, a single API client file, one deploy path. Any developer, and any coding agent, already knows those tools.

Before the builder leaves, have their agent write the handover note while the context is still on their machine:

```prompt
Write HANDOVER.md at the root of this repository for a developer who has never seen it. Cover: how to install and run it locally (exact commands from package.json), every environment variable and where its value lives (Cloudflare build variables, runtime secrets, .env.example), which Diggama blueprints the code reads and which file reads each one, how a publish in Diggama triggers a rebuild, how to deploy by hand, and the three changes most likely to be needed next. Read the code to get each answer; do not guess. List anything you could not confirm from the repository at the end, as questions for me.
```

## How locked in are you with each option?

Framer locks in the most: it [does not offer HTML export](https://www.framer.com/help/articles/can-i-export-my-website-to-html-and-self-host-it/) for self-hosting, because its sites depend on platform services. Webflow exports static pages as HTML and CSS, but not CMS pages, and forms stop working. The coded stack locks in the least: the code is yours, and the CMS content leaves through an API.

Webflow's own [pricing FAQ](https://webflow.com/pricing) is explicit about the limits: code export needs a paid Workspace plan, dynamic content "must be exported on a collection-by-collection basis", and exported code cannot be imported back. In practice, leaving Webflow means rebuilding the design and moving the content separately, which the [Webflow migration guide](/guides/migrate-from-webflow/) walks through.

On the coded stack, the lock-in moves to the CMS, so check its exit before you commit. Diggama's [export endpoint](https://docs.diggama.com/export) returns up to 10,000 records of a blueprint as JSON in one request, and the regular API reads everything else. The frontend does not care which CMS sits behind `src/lib/diggama.ts`: switching means rewriting that one file and moving the data. Hosting is portable too: the prerendered pages run on any static host, and only the small form endpoint is written for Cloudflare's runtime.

## Which one makes the fastest site?

All three can serve a fast site from a global CDN. Webflow [states](https://webflow.com/pricing) that its hosting uses Cloudflare's CDN, Framer's Pro plan serves from more than 300 locations (20 on Basic), and the coded stack runs on Cloudflare's network directly. The real difference is control over what each page ships.

On Webflow and Framer, the platform decides the runtime and you optimize within it. On the coded stack, you control every byte: a prerendered Astro page with no client-side JavaScript is hard to beat, but your agent will also add a carousel library if you ask, so performance becomes your responsibility. Run the [website grader](/tools/website-grader/) on the live URL after each significant change, whichever option you pick.

## How does AI fit into each option in 2026?

All three now work with AI agents, but at different layers. Webflow and Framer put AI inside the platform, to generate sections and content on their canvas. The coded stack puts AI in charge of the whole codebase, with the CMS exposing content to it through MCP.

- **Webflow** includes Webflow AI on every site plan and has an [MCP server](https://developers.webflow.com/data/docs/ai-tools) that lets AI clients such as Claude, Cursor or Windsurf create and update CMS items and modify elements and styles on the canvas. The CMS tools work without Webflow open; the canvas tools run through a Designer companion app that must stay open.
- **Framer** includes AI credits on paid plans for its agents, and lists "external agents" that connect your own tools, free during a preview.
- **The coded stack** uses any coding agent for the code and the CMS's MCP server for the content. Diggama's MCP server lets your agent read the content model, draft records and translate them, limited to the abilities of the token you create under Configuration › Connect to AI: a read-only connection is never offered a write tool.

The Diggama server authenticates with a Bearer token, which Claude Code, Codex, Cursor, Gemini CLI and Copilot in VS Code all send from their MCP config. Chat apps are more limited: ChatGPT on the web [cannot present a custom API key](https://developers.openai.com/plugins/build/auth), and claude.ai and Claude Desktop accept one only through [request headers](https://claude.com/docs/connectors/custom/add-unlisted#authenticate-with-request-headers), a beta open to a limited set of organizations. The [MCP server guide](/guides/headless-cms-mcp-server/) covers each client's setup.

## Webflow vs Claude Code, Codex or Cursor: is it worth leaving Webflow?

Leaving Webflow is worth it when you are paying for things Webflow prices per unit (editor seats, locales, bandwidth tiers) or fighting its limits (custom logic, integrations, page structure), and you have someone who can run an AI coding agent and review its work. It is not worth it if your Webflow site works, your designers own it, and nobody wants to maintain a repository.

Signs that leaving makes sense:

- You already write custom code embeds to work around Webflow, and they are the hardest part of the site to maintain.
- You need features a static site builder does not do well: logged-in areas, complex forms, search over your own data, integrations with internal systems.
- Your traffic or language count pushes you into the expensive bandwidth or Localize tiers.
- A developer is already part of the team, and they would rather work in code than in the Designer.

Signs that you should stay:

- The site is a marketing site, it changes weekly, and designers make those changes visually.
- Nobody on the team will read a diff or fix a failed build.
- Your three-year subscription total is a few thousand dollars and nobody is complaining about it. A migration costs developer days, and the [cost calculator](/tools/website-cost-calculator/) will tell you how long it takes to pay back.

There is also a middle path. [Webflow Cloud](https://developers.webflow.com/webflow-cloud/intro) hosts Next.js, Astro and Vite apps next to a Webflow site, on a subpath of the same domain, so you can keep marketing pages in the Designer and have your agent write only the parts Webflow cannot do (a calculator, a customer area). Its request, CPU and storage limits come with the site plan. It suits a site that is mostly marketing pages with one or two pieces of custom logic; it does not remove seat, locale or bandwidth pricing.

The same logic applies to Framer, with one difference: since Framer has no code export, leaving it is always a rebuild, so the case for leaving has to be stronger.

## Which should you choose?

Choose by who will change the site most often. If it is a designer, choose a visual builder. If it is a developer working with an AI coding agent, with marketers editing content on the side, choose the coded stack.

**Choose Framer if:**

- The site is design-led: a launch page, a startup's marketing site, a portfolio.
- The team is designers and marketers, with no developer.
- You want the lowest subscription cost for a small site, and speed of iteration matters more than control.

**Choose Webflow if:**

- The site has a real content structure (many collections, references, hundreds of pages) and a designer who knows CSS.
- Editors need on-page editing and the team will not maintain code.
- You need built-in localization with AI translation, managed in the same tool.

**Choose the coded stack (an AI coding agent, a headless CMS, Cloudflare) if:**

- Someone on the team can run a coding agent such as Claude Code, Codex or Cursor, review its changes and fix a build.
- You want to own the code and be able to move it to any host.
- The site needs logic a builder cannot provide, or your costs are driven by seats, locales or traffic.
- Marketers need to edit content without depending on the developer for every change, which is exactly the job of the CMS. The [headless CMS comparison](/guides/best-headless-cms-2026/) covers the options beyond Diggama.

## How do you try the coded stack before you commit?

Build one real page on the coded stack, with real content in a CMS, and have one editor publish a change through it. A pilot of a day or two tells you more about the workflow than any comparison table.

1. Rebuild your homepage and blog with your coding agent, following the [pillar guide](/guides/build-a-website-with-ai/) or a framework tutorial ([Astro](/guides/build-a-website-with-ai/astro/), [Next.js](/guides/build-a-website-with-ai/nextjs/)).
2. Put the content in a CMS. Diggama's trial runs three months without a credit card ([sign up](https://manage.diggama.com/register)).
3. Deploy to a Cloudflare preview URL and ask a marketer to change a headline and publish a post without your help.
4. Compare the result, and the time it took, with your current site.

Webflow's code export needs a paid Workspace (Core, $19 a month billed yearly). On the free Starter Workspace, give your agent the live URLs and screenshots instead; CMS collections can still be [exported as CSV](https://webflow.com/feature/export-content-as-a-csv), one collection at a time, from the Designer. Image fields come out as URLs on Webflow's servers, so copy the images before you close the Webflow site. Use this prompt in a fresh repository to scope the pilot before writing any code:

```prompt
I am comparing my current Webflow site with a rebuild in Astro, content in the Diggama headless CMS, and hosting on Cloudflare Workers. The Webflow code export is in ./webflow-export (if that folder does not exist, fetch the live pages listed in ./urls.txt instead), the CSV export of each CMS collection is in ./cms-csv, and screenshots of the homepage and one blog post are in ./screenshots. Read them and write PILOT.md with: the list of pages and their URLs, the Diggama blueprints the homepage and blog need (one table per blueprint, with field names and Diggama field types such as text, rich-text, image, slug and reference), the parts of the site that use Webflow interactions or custom code embeds and how you would rebuild each, and anything you think will be hard to reproduce. Do not write application code yet.
```

If your team prefers the canvas, you have spent two days learning that Webflow or Framer is the right tool. If editors publish happily from the CMS, the [migration guide](/guides/migrate-from-webflow/) covers the rest: content export, images, redirects and the DNS switch.

## FAQ

### Is building with Claude Code or another AI agent cheaper than Webflow?

Not on subscriptions for a small site. Among the usual paid plans for the main agents, the cheapest is Claude Pro at $200 a year billed annually; ChatGPT Plus (which includes Codex) and Cursor Pro cost $20 a month. A headless CMS adds its own fee, while Webflow's Premium site plan, the one with the CMS, is $25 a month billed yearly. The coded stack becomes cheaper when you would otherwise pay for many editor seats, extra locales or high bandwidth tiers, because Cloudflare does not charge for static asset requests and a CMS plan covers a fixed number of users.

### Can a non-developer maintain a website built with an AI coding agent?

They can edit its content if the content lives in a CMS, because editors then work in a dashboard and never touch the code. Changing the layout, fixing a failed build or upgrading the framework still means running the agent (Claude Code, Codex, Cursor or another) and judging its changes. If nobody on the team is willing to do that, a visual builder like Webflow or Framer is the safer choice.

### Which AI coding agent should I use to build a website?

Any of the main ones fits this stack. Claude Code, OpenAI Codex and Cursor all work on a local codebase, follow a project instructions file and connect to a remote MCP server with a Bearer token, so the choice comes down to the subscription you already pay for and the editor you prefer. Their usual paid plans each cost about $20 a month; GitHub Copilot Pro ($10) and Gemini CLI's free tier cost less.

### Is Framer better than Webflow?

Framer is faster for design-led marketing sites and cheaper at entry level: Basic is $10 a month and Pro $30 a month, billed yearly. Webflow gives more control over structure and CSS, a larger CMS (20,000 items on its Premium plan) and a static HTML and CSS export on paid Workspace plans. Pick Framer for speed of design, Webflow for complex structure and content.

### Can I export my site from Framer or Webflow and host it elsewhere?

Webflow exports static pages as HTML and CSS on paid Workspace plans, but CMS items and pages are not included and forms stop working. Framer states that it does not offer HTML export for self-hosting, because its sites depend on platform services. In both cases, plan to rebuild the site rather than move it.

### Do I still need a CMS if an AI agent writes my website?

Only if people other than the developer edit content. With copy in the code, every typo fix becomes a prompt, a diff review and a deploy. A headless CMS gives editors a form, drafts, roles and a publish button, while the agent keeps working on the code and can read the content model through the CMS's MCP server.

### What happens to an AI-built website when the person who built it leaves?

The site keeps running, because it is static files on a host. The risk is the next change: someone has to find the repository, the deploy settings and the tokens. A README, an AGENTS.md file with the project's rules, access to Git, Cloudflare and the CMS held by the company rather than a person, and a standard framework make the handover a day's work instead of a rebuild.