In this application posted by Zero 2 Hero on youtube Dominic is mentioning web 3:
"Lastly of course internet computer natively supports Web 3, Caffeine actually doesnt yet, the functionality is there, it hasnt yet been switched on, but you are going to be able to create multi-chain applications, that process tokens and interact with smart contracts across multiple chains that are fully trustless, just by talking to AI.
Uhm, that will be coming within two to three months"
Open Questions:
1: How will this look and what chains will be supported
2: Does this mean you can then use backend wallet functionality within your app
3: What restrictions will remain / be built in (assuming it will not be a free for all)
4: Can users now more easily mint their own Tokens and NFT on ICP and other blockchains?
5: Does this mean people will now create their own AI controlled wallets
6: Could an application be allowed to have only either or both fiat and crypto integration
7: Will Caffeine AI (in the future) integrate with SNS
That’s a fair point, and it does make sense. Still, a lot will depend on the limitations in place and how quickly these features continue to improve.
In my experience testing Web2 apps, I already run into quite a few bugs. Because of that, building a full DeFi application with Caffeine AI feels pretty challenging right now. Even something that sounds simple — like minting a token or NFT on Ethereum — comes with additional layers of complexity that aren’t trivial to handle.
I’m also curious whether there will be built-in functionality, templates, or engines to help users deploy these kinds of applications more easily. Other platforms, like Anthropic’s tools, already support this fairly well, though they operate in environments with far fewer restrictions than what users currently face in Caffeine.
The 2–3 month timeframe also felt loosely defined. I worry the rollout may end up being incremental, with ongoing limitations, while potential users continue choosing more established and flexible alternatives.
Automated (non-AI) development tools for DeFi and Web3 have been around since at least 2017, so there’s already a significant head start in the ecosystem. Builders tend to gravitate toward environments where they can fully use their creativity without heavy constraints.
On top of that, developing complex applications requires substantial processing power. I’m not sure how that’s being handled here, but it does feel like Caffeine is somewhat held back — especially compared to other systems where users can directly configure the compute resources I need.
Very few people will with some exceptions will spend weeks prompting Caffeine and think 500 prompts is acceptable, whereas it could have been done with the same engine and a different wrapper with 2-3 large prompts followed by huge amounts of small prompts where you make tiny incremental improvements.
I think they are already “on” it’s just that the guard rails are extremely strict.
you can get the backend to talk to the ledger (with very careful specific prompts that have 0 possibility of funds being “stolen”), but if you try to add an option for the admin for example to arbitrarily withdraw balance from the canister, it won’t code it.
Thanks — I was actually referring to you in my previous comment, lol. @WebTreeSoftwareSolut
Regarding the guardrails you mentioned, I do understand why they’re in place. The product is still in beta, and there’s also a strong focus on orthogonal persistence, which in theory should make the system more resilient and reliable.
That said, as a builder, I’m not a big fan of those limitations. I tend to work on fairly complex applications, and I prefer having the freedom to build without too many restrictions getting in the way.
I would prefer a toogle switch where you have caffeine unleashed without restrictions.
Whenever I want to improve one line of codes, I do not want to have to wait for five minutes while Caffeine is rewritting the entire app which must be a huge waste of compute power.
Some of your posts shows a level of technical understanding which goes way above and beyond the typical user which would benefit from using Caffeine AI at the moment
What is the reason you are spending so much time on testing?
because I am only one person, and this thing multiplies my productivity by 200x
I would say a lot of my understanding is surface level. I understand main ideas, design and architecture patterns, But not specific implementation details.
I can read the code and know what’s going on, but that’s not the same as writing it.
I really should bite the bullet and actually learn rust though… thats probably holding me back.
I am about to start a boss fight! Lord Jesus help me
@WebTreeSoftwareSolut
You’re clearly one of the most advanced Caffeine builders out there right now — the apps you’ve shared are seriously impressive, objectively compared to other Caffeine Apps.
But I was curious about your comment that this multiplies your productivity by 200x. What are you comparing that to exactly?
While I am willing to concede you may be one of the greatest coder in your vertical, I believe as a novice I can outdo you quite easily as long as I get to use the same engine as you.
Perhaps a fun adventure if you do not have anything to do for the next couple of hours?
unfortunately I have to go do my real job, Although I constantly watch this forum, I assure you, I am actually employed in the real world. I am curious though what you have in mind. dm me.
I am comparing it to writing the code myself. And deploying via dfx cli, which is what I did before they publicly released caffeine.
I understand your priorities
I am not working today but maybe another day
I was thinking somewhere along like this:
You can set the terms but I suggest:
Time:
Three - five hours App Idea:
You decide the app idea Languages:
Since you are restricted to Motoko perhaps you can choose either Motoko or Rust for me (although I prefer Rust) Engine:
Has to be anthropic models Criteria’s for winning:
Best / Biggest / Most Beautiful code with operational functionality. Can be decided by voting, or we can just agree between ourselves, I would have no problems concede to loosing, in fact I would love for you to win.
It could be great fun and maybe that data is worth something to someone
