BIG NEWS: Elements 3.0 is out of beta! šŸŽ‰

Was going to try this later today with size edits:
(Computer is currently rendering video)

Work on the Elements project that is open now, using the Elements MCP tools.

Site brief: [site name] for [audience]. Look and feel: [3 adjectives or 2–3 reference sites].

Rules for the whole job:

- Keep every page, page name, and URL. Keep the existing text; restyle and rearrange it rather than rewriting it.

- Build only with Elements’ built-in components and templates. List the available templates before choosing.

- Design mobile-first: phone, then 768px, then 1024px and up. First confirm the theme’s 768 and 1024 breakpoints are enabled.

- Use the theme’s existing colors and fonts unless I approve a change.

- Don’t edit CMS collections or posts. Never publish.

- Save the design plan and every decision to AI Memory.

- Stop at the end of each phase and wait for my OK.

Phase 1 – Audit, no changes: list each page with its sections, current title, meta description, and any small-screen layout problems.

Phase 2 – Plan: propose the type scale, spacing, section order for each page, and button style.

Phase 3 – Rebuild the Home page to the plan.

Phase 4 – Rebuild the other pages, one page per step.

Phase 5 – SEO: for each page, write a unique browser title of 50–60 characters ending with the site name, a meta description of 150–160 characters, and matching Open Graph title and description. Show them in a table before saving.

Phase 6 – Consistency pass: one H1 per page, even spacing, matching buttons. List every change.

Start with Phase 1.

I wish we could take all the credit, but this is down to the difference in model, not something we’ve done specifically for Claude.

We have done a lot of work on the ā€œElements Harnessā€ so Elements will work nicely with any AI model. However, the better the model, the better the results! I’ve found local models to be not that great designing and building sites, for best results use the frontier models from the big AI labs, Grok, Anthropic, and OpenAI.

This is great. I connected with MCP with Claude Desktop. I told it I had some custom components in my own Dev Pak. I went through and record the project to notes and in the process found I had a custom PDF Viewer component on a test page that wasn’t completed or part of my Dev Pak. We went through and made it a component on my Dev Pak. Updated the test page made a few tweaks and it is now functional on my test page. So when I decide where I want to place it on the site I can put it on any page. It can load a single PDF dropped in the well or a folder of PDF and it builds its own menu buttons to select.

This is really cool!!! Dan & Team excellent work…

I’ve just posted a video on the difference between the Elements AI Assistant and the Elements MCP server. Hopefully this will help users that are newer to AI get started.

Let me know if you have any further questions :slightly_smiling_face:

Thats great Dan

I have a paid version of Claude so don’t really know. I just use my credits as and when.

In terms of what I was able to do not that much really but I was checking SEO, duplicating pages and content and re designing menus. I will try and keep a closer eye on it to see how many tokens it eats up in the future.

To be fair for the sake or £15 for Claude a month its worth just trying out to see how much time it saves you. If it saves you a lot of hours its going to be worth it.

THANKS I like to play with my time as I am retired and it gives my brain something to do. I see AI as an assistant when I get stuck or just get frustrated trying for a period and can’t do want I want it to do. Sadly I dont have Ā£15 as my wife keeps going shopping1 :slight_smile: :slight_smile: :slight_smile:

Still working through all my models, now on my best AI-capable Mac. There’s definitely differences. The top Gemma (Gemini from Google) models seem to do okay, but Claude still seems the best at doing real design work. All the models I’ve tested so far, included QWEN seem to be capable of making a tweak type of change, as you suggest. Only Claude seems to be reading the bigger picture better.

It showed up when I tried to enter an API key, but I see in recent posts that you have figured it out.

Hi Thom,

What Mac specs do you have, what amount of unified memory ?

For local models, 27B-35B is the real starting point, while 70B-125B is where things get even further interesting. Give Qwen 3.8 27B or Qwen 3.8 Flash-Next a try if you have the spec capacity.

All my Macs currently have 64GB unified memory. However I do have a lot of things running simultaneously at the moment. I need to finish a few projects before I’d have a clean 64GB to experiment with.