I watched Jerry’s latest video just now about Claude Code and ICP. He was commenting on how powerful it is. I currently run a Wordpress/WooCommerce website that is connected to one of the hosted online course platforms like Teachable but not that one.. It would certainly be great to be able to get off Wordpress and not have to worry about dealing with core and plugin updates on a near daily basis. How feasible is it today to actually be able to re-create (a better version of) my website, along with the features of the online course platform that I need, Stripe/Paypal enabled, all on ICP via AI? If it’s all command line controlled, how do you implement things like photos, product images all all the normal shopping cart type stuff?
Can’t you try out replicating your site on caffeine
I don’t know.. I recall Jerry saying he tried Caffeine, but did better with Claude Code connected to ICP, however that works..
I think your project would benefit from AI-assisted development. One thing I would think about is whether you actually need to recreate something like WooCommerce, or whether you need to identify the specific business capabilities your website requires.
WooCommerce is extremely powerful because it has accumulated many years of features, but many businesses only use a portion of those capabilities.
A different approach could be to use Claude Code (or similar AI coding agents) to build a focused ICP-native application designed specifically around your business requirements.
The development workflow is changing. It is becoming less:
“Learn every framework, write every line of code, build everything manually.”
And more:
“Understand the problem, design the architecture, let AI help implement, then review and accept the result.”
Claude Code is interesting because it is not just a website generator. It can work with an actual codebase. It can:
-
inspect existing files;
-
understand project structure;
-
create and modify code;
-
debug errors;
-
run tests;
-
help implement features.
The human still needs to provide the important parts:
-
business requirements;
-
architecture decisions;
-
security boundaries;
-
acceptance criteria.
The recent WordPress-on-ICP discussion is also very relevant here. There is an interesting effort to run WordPress, PHP 8.5, SQL, and WooCommerce workloads fully on ICP:
WordPress running 100% on-chain on the Internet Computer with PHP 8.5, WASQL and WooCommerce
If this type of technology matures, it could become an important migration path because existing Web2 applications could move to ICP without requiring a complete rewrite.
However, I think there are two different strategies.
Option 1: Move existing WordPress/WooCommerce applications to ICP
Advantages:
-
Existing ecosystem
-
Familiar tools
-
Thousands of plugins
-
Faster migration
Challenges:
-
You inherit WordPress complexity
-
Plugin compatibility becomes a major issue
-
Performance and cost need to be proven at scale
Option 2: Build an ICP-native application
Advantages:
-
Cleaner architecture
-
More customization
-
Better use of ICP features
-
Easier to design exactly what you need
Challenges:
- You need to build more pieces yourself
For a specific business website, I would probably lean toward the second option. You may only need 10–20% of WooCommerce functionality, and AI coding agents make building that custom functionality much easier than it was before.
The comparison with Caffeine AI is also interesting.
I would think of Caffeine as more of an AI application builder. It is excellent for quickly creating prototypes and simpler applications.
Claude Code is closer to an AI software engineer. It gives you more flexibility, but it also requires clearer requirements and more architectural direction.
A simple comparison:
Caffeine:
“Help me create an application.”
Claude Code:
“Help me design, implement, debug, and evolve my application.”
For simple applications, a builder approach may be enough.
For more complex applications involving:
-
custom workflows;
-
identity;
-
payments;
-
AI agents;
-
unique business logic;
I think the coding-agent approach has a lot of potential.
The bigger shift is that developers may not need to manually write every line of code anymore. The valuable skills become:
-
understanding the business problem;
-
designing the architecture;
-
making good tradeoffs;
-
reviewing the implementation.
This is actually very aligned with where ICP seems to be heading: making it easier to build serious applications without requiring every developer to code and just leverage AI.
@alfonso Thanks, that is really good insight and explanation. And no, I wouldn’t want or need to re-create Woo and all its features. I would just want to create the features and workflows that I need like you said. One of the issues that I can see now is video hosting. By using a hosted online course platform, they have an enterprise api type account with Vimeo, so I’d lose that and would have to see about re-creating, or finding another video hosting solution that would work on ICP/Claude.
Another thing I read about Caffeine, is at least for now, it only supports Internet Identity logins, which certainly isn’t applicable to the masses like Google, or Facebook login.
I think the Vimeo question is a really good one because it gets into the difference between “hosting an application on ICP” and “having to rebuild every external service that the application uses.”
You do not necessarily need to recreate Vimeo or move all video hosting onto ICP.
An ICP-based website can still integrate with external services through APIs. The fact that your website runs on ICP does not mean it is completely isolated from the outside world.
A possible architecture would look something like:
User
↓
ICP website / canister
↓
Course management logic
↓
Vimeo API
The ICP application could manage things like:
-
courses;
-
users;
-
subscriptions;
-
permissions;
-
lesson progress;
-
certificates;
-
AI assistance;
-
personalized learning history.
Vimeo would continue handling the things it is already very good at:
-
video storage;
-
encoding;
-
adaptive streaming;
-
bandwidth delivery;
-
video playback.
This is actually how many modern applications are built. They do not try to replace every specialized infrastructure provider. They own the parts that create their unique value and integrate with services that solve specialized problems.
For example:
A course platform does not need to become a video company.
It needs to provide the best learning experience.
The important data and workflows could live on ICP, while the heavy video delivery could remain with a specialized provider.
The same idea applies to authentication. Internet Identity is a very interesting ICP-native option, but using Claude Code is different from using a platform like Caffeine. Claude Code is an AI coding agent, so you are not limited to only one login method.
You could build a user experience such as:
-
Google login;
-
Apple login;
-
email login;
-
Internet Identity;
and connect those users to your ICP application.
The key difference is:
Caffeine is more like:
“Describe an application and have the platform generate it.”
Claude Code is more like:
“Have an AI software engineer help you build the application you design.”
So with Claude Code, you have more architectural freedom. You decide whether users authenticate through Google, Internet Identity, or a combination.
I think the larger design question is not “How do I recreate everything that WordPress/WooCommerce/Vimeo already do?”
It is:
“What parts of the experience are unique enough that they should become my application’s core?”
For an online learning platform, the differentiator is probably not the video player. Video is becoming commodity infrastructure.
The differentiator could be:
-
remembering a student’s history;
-
personalized learning paths;
-
AI assistance;
-
progress tracking;
-
community;
-
long-term relationships with learners.
Those are the areas where an ICP-based architecture could become really interesting.