Mere Morpheus

Mere Morpheus artwork

Dappit's AI Aggregator Suite | Personal AI Meets Web3

54 minAudio

Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via their podcast app and to have them all in one place. If you'd like to participate they are held weekly on the @MorpheusAIs Twitter spaces. So enjoy Episode #61! Recorded: 5 February 2026 AI SUMMARY: This week's builders update features Achilles from Dappit who discusses the rise of agentic AI colliding with web3. Dappit is an "AI aggregator" for building decentralised apps end-to-end from requirements gathering through design coding with Opus 4.5 smart contract generation auditing and deployment. Achilles explains the real bottleneck is not model intelligence but communication context and orchestration arguing that smaller specialised open-source models with agentic workflows may outpace the bigger model race. The conversation covers the practical limits of today's models for web3 and Dappit's training of a web3-tuned open-source model. It zooms out to the cultural moment around agents security pitfalls of unsecured local installs and the case for personal fiduciary agents aligned via smart-contract rails. Topics include agentic payments on-chain reputation trustless coordination and the future of "software that builds software" debating alignment justice for autonomous systems and AI's impact on human jobs. Timeline: (00:00:00) Intro (00:01:14) Guest intro: Achilles on Dappit as an AI aggregator for web3 (00:02:24) How Dappit works: models, browser app, and workflow overview (00:07:26) Open-source vs proprietary models: Kimi, Sonnet, and closing the gap (00:11:18) Diminishing returns, context over size, and agentic collaboration (00:15:09) Costs, routing to open models, and platform sustainability (00:18:16) Claude/Moltbot moment: why agents just went mainstream (00:22:45) Security first: sandboxing, Agent Zero, and safe setups (00:26:24) Agent economies: trustless coordination, x402, and on-chain reputation (00:28:42) Sci-fi parallels: Altered Carbon and the death of one-size software (00:31:00) Personal super-agents and owning your intelligence (00:33:10) Alignment to the individual and free-market dynamics (00:37:01) Justice for agents: disputes, subjectivity, and decentralised norms (00:41:34) Personal firewalls and fiduciary agents vs ad-driven platforms (00:44:42) Model races and the GPT-4o saga: hype, cults, and sentience claims (00:47:01) Agents hiring humans: recursion, labour creation, and new jobs (00:51:02) Intent matters: values, goals, and briefing your personal agent (00:52:31) Wrap-up and next steps: Morpheus 26 DAPT collaboration, MCP server If you would like to support me for the value I provide you can use the referral link below when depositing capital and we will both earn bonus MOR. Alternatively you could send me a token of appreciation directly to my wallet ;) Referral Link: https://dashboard.mor.org/capital?referrer=0xE935f231c99c04Ee0b4532a3d0BdA81B152a0384 Wallet Address: 0xE935f231c99c04Ee0b4532a3d0BdA81B152a0384 Connect with Mere Mortals: Website: https://www.meremortalspodcasts.com/ Discord: https://discord.gg/dX9BEhT3kz Twitter/X: https://twitter.com/meremortalspods Instagram: https://www.instagram.com/meremortalspodcasts/ TikTok: https://www.tiktok.com/@meremortalspodcasts Value 4 Value Support: Website: https://www.meremortalspodcasts.com/support Paypal: https://www.paypal.com/paypalme/meremortalspodcast

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

David JohnstonIn the meantime, I wanna say it's been a wild week for AI. I hope you guys have gotten all your open clause set up and managed to access the open source. There's a rumor going around that, the founder is gonna go, work for Meta, so we'll see what that means for the open source. Hopefully, they'll keep contributing, but it's amazing to see sort of this breakout moment for

smart agents. Personal AI, hosted, acting as coordinator, accessing all these other tools, really sort of, great validation of what we've been talking about the last three years.

Bowen BelmerAbsolutely. I would say, David, that, like, everything that's happened over the past two weeks with Moldbot has been an amazing validation for, our community's view of the future. Like, what does that look like? Okay. I just dropped a note. I was just dropping a note in the in the Venice Discord as well, given the amount of overlap in community support between the two projects. Cool. Well, let's dive into it. So joining us today, we have Achilles

from DAPT, who I will let you give a proper introduction as to kind of who you are and what DAPT is, because I know you've had some changes. So I wanna make sure, you know, I I don't accidentally misspeak.

DappitI appreciate that. Thanks for having me, guys. Yeah. Of course. I'm Achilles. I do things in crypto. DAPT Rayhan, it's great to see you, man. DAPT is we we're calling it the first AI aggregator. Right? So we believe that that software in its current form is kind of on the way out. And there's, you know, the vibe coding thing you guys were just mentioning, Moldbook and Cloudbot and all that kind of stuff. AIs are gonna be our main access to the Internet going forward. So we figured, why not create a platform where everything that you wanna get done in web three

can be done with an AI aid? So we do everything from design to deployment to smart contract creation, auditing, vibe coding of the actual software, whole nine yards from start to finish is DAP decision.

Bowen BelmerCan you talk about how you guys are actually accomplishing that? So, like like, in terms of, you know, models used, where are you sourcing inference, and, like, how in in in practical terms, if somebody wants to connect with DAP, like, what is actually happening on the back end?

DappitOkay. So in practical terms, if if you're you're using it through your web browser. Right? We have an MCP server coming out of near future, but that's currently not live. So you're just it's an application that you're utilizing online. How it works is the difficulty in actually getting functioning software out of AI is not the AI's ability anymore. It's our ability to communicate what we want and to describe, like, cohesive functions. Right? So we've taken a few approaches to making this a little bit easier for the user. First off, DAPT decoder is very web three tuned.

It's our system config is, you know, like, 100 lines long. It's one of the cooler things we've been able to accomplish. In the future, we're gonna migrate that to an actually fine tune LoRa model. But right now, DAPID is Opus 4.5 because it's the only thing that can handle it as the model get a little bit better. I know I was talking with, you know, one of you guys about Kimi 2.5, and she's doing great. We're getting there. Yeah. Who's that? Yeah. Yeah, man. Yeah. Yeah. Slowly but surely. Right? We'll be able to do this open source. I've actually experimented a lot with attempting to do this

with an array of models on a local system. I'm not successful yet. So as they get better, we'll get better. But DAPID starts with AI called Ade, and she's a spin off of Haiku. What she does, she's trained to pretty much just ask you questions, and she has web three context. So she asks you questions about what you wanna build. She makes sure she gets all the details as to what that's gonna look like, and then she engineers a prompt. That also gets adapted into your white paper and your pitch deck. And this is all for version two, which if you're a beta tester, we love you guys. That'll be coming out in the near future for the public. Right now, you guys are just using DAPT.

So Ade takes the context of your product and she breaks it into the different areas you're gonna be building in. Right? So a white paper, pitch deck, you know, a Gitbook, read me file, all the things that a project needs. Then you move on to the design aspect. Design is making or uploading images, making sure the the theme is exactly what you want, making sure it looks how you're you're looking to go to market. Right? Then we move on to coding. The coding aspect is DAPIT, which is Opus 4.5.

He, you know, puts together I say, hey. I was just talking with someone about this. I really anthropomorphize these models after spending so much time with them. He puts together the code and DAP aspect and the smart contract aspect if that is necessary. Now we have two ways of approaching a smart contract. We currently have a custom manner that I I guess I won't talk about on spaces, but it's almost guaranteed, like, a 100% accuracy. The smart contract that's selected that works in the system will work with your DAP because your DAP will work around the smart contract.

Then for real custom builds, we have smart contract generation, which is where that MCP server that we've discussed talks about. Web three and AI are coming up, but they're not exactly where we want them to be yet. Web three is a pretty esoteric space. Right? So even a model like Opus, it'll know solidity, but it won't know how to optimize gas and things like that. So we have to have a few workarounds for that. If you are generating a custom smart contract, then you move on to the audit function. Make sure it's safe. Right? Make sure it's gonna actually, you know, serve the purposes that you're looking for, and then you deploy it. You deploy the smart contract that is. The smart contract address and all necessary functions get imported back into the software,

and now you have a working DApp. Then what do you do with the DApp? This is where publishing and launching whatever your project is comes into play. We currently have integration with things like for cell, Netlify, GitHub, Supabase. So any, you know, databases you need in the back end are are accounted for and that we can build those as long as you plug in the right API keys. But in the future, we're gonna be adopting a little bit more of a web three centric focus where, you know, you'll be able to go directly to, or

or, you know, whatever non smart contract based deployment platforms that you want to put your project on. Because we don't wanna limit web free builders. Right? Some people wanna make a full custom smart contract that has functionality and gets called, and some people just wanted to plan with cool meme. Whatever you guys want. It's all built in. And then, of course, things like, you know, Reddit and publicizing it are all also automated. And, yeah, I think that's a pretty thorough description of our current platform.

Bowen BelmerYeah. Thank you. No. That's super helpful. I feel like I have a much better understanding now. As as a nontechnical person, I feel like I have a much better understanding now of kinda how how the whole stack works, and, you know, what the goals, inputs, and outputs are. Can you speak a little bit, Achilles, to what parts, especially on the AI side of things so so obviously, with the focus being dApps and, like, building web three tools and products and and projects, really, That part is obviously

tuned to, like, the decentralized future that that probably everybody on this call is partial to and and building towards. But on the AI side, I know you said that right now it's running Opus, and and there had been talks with community members within Morpheus about something like Kimi. What I guess, what are the restrictions? Like, what what are the things that you would need in order to have an open source, and, like, decentralized version on the AI side of things? Like, where are those limitations right now? What does that look like?

Kyle (Bow Tied Blue Fin)And, and be before you get into it, so that that was me. This this is Kyle, by the way. Oh, Kyle. What's up, man? Yeah. Yeah. We we spoke a lot about and and we've been speaking a lot about different models. And I think that the space right now, and I really wanna get your opinion, but I I want to set the stage a little bit, that there are different models for different purposes. Right? And and there's kind of a sliding scale. Right? We're not gonna use you know, you you talked about haiku or fine tune of haiku, for example. You're not gonna use opus level intelligence for something that you only need haiku on. Right? But there and and all of the different steps in the middle depending on what the application is in which you're using the AI.

And I would agree right now that OPIS is the best coding model out there. Right? I'm not like, that that's my opinion. That's the the world's opinion. And there are some cases where that's needed, right, for for what you're trying to develop. I think that I'm gonna make two points, and then I I want I want your opinion on it. I think that we can't just look at where is, like, where's OPUS five then and where's OPUS six and where's OPUS 35

and say that, you know, the top proprietary model is always the benchmark. Right? Sometimes it may be. But I think that where we're at with OPUS 4.5 right now is kind of the benchmark of where can you build ultra high levels of code. And I think that we're very close to the open source models getting to that benchmark. Right? Kimi 2.5, I think, is an incredible model. I think it's at the level of you you know and and we can debate, but about at the level of SONNET four or five.

And I think that the next version of these models that'll come out, are going to be around the levels of OPUS. Right? Even if OPUS six is a little bit better, I think that we'll be at the point that we can run, you know, our platforms through open source code at a very competent level in, you know,

David Johnstonthe coming months, I I think we're we're And we we sure didn't have to wait long. I saw thirty one minutes ago, they dropped Opus 4.6. Did Sonify come out as well? No. They haven't said anything yet. People are speculating they might suck the juice out of 4.6 for a week and then do Sonnet five next week, but who knows? It's sort of just speculation. But, yeah, people were expecting some drop today. I guess it's this Opus 4.6. But I agree with Kyle. The open the open source models aren't far behind,

and I I expect fully will end up with an open source model within the next few weeks, maybe a month that is 4.5 level.

DappitWell, I'd be sure I have an analogy that might make sense to you guys. I mean, you play video games here? Yep. Yeah. You heard of the law of diminishing polygons?

David JohnstonSure. You you had a few at the beginning. It looks really like a huge improvement. You keep going up that scale and it gets harder and harder to make improvements.

DappitRight. So we don't need a better coder. We need a coder with more context. Right? We need to be able to absorb more data and absorb that data better. We need agentic distribution platforms, which I think any developer in the space is currently trying to figure out. I know one that's allegedly cracked it, but we're we're all trying to figure out how to make agents coherently code in the same space separately. So you don't have to use massive models. Right? You have 10 GPT four point o's accomplishing what two Opuses could do instead of,

yeah, a giant Opus model. So it's not necessarily that the models currently are needing to be improved, at least not for current necessities. It's that they need to be optimized. How do I put this? So in the web three example, right, Opus can do all of the coding just fine. It's not great for the web three side of things because there's no model that's really great for this the web three side of things. So rather than just give Opus more context,

my approach is to take a smaller open source model that I can actually, like, Laura train and give it the context to work with Opus. So I'm currently training a I can never remember these model names, the open source ones. It's a QUEN three thirty b dash a three b. Yes. And it's what's working name is Big Achilles right now. We're training that. We're harvesting web three data. Right? We've got scrapers running on CoinGecko and DexScreener's API and pulling whatever's however slow their API rate is due at alchemy

and harvesting as much real usable data that we can train this thing on so that it can weigh in on the build. Right? It can write the smart contract. It can do so Opus won't be necessary for that aspect of the building longer. Then, you know, honestly, Sonic will probably do just five, just five in its place. I don't think the bigger, better, more billion dollar training models is a sustainable method. That's why I brought up the polygons. Right? Like, I still play the Witcher three. It's I don't know however many years old. It's a phenomenal game, and it looks beautiful.

Right? Cyberpunk looks pretty cool, but, like, witcher three still looks awesome. There's no reason for us to to sort of toss that technology away. That's, yeah, that's how I feel about the current set of AIs. We're gonna start using better models for better actions as opposed to the catchall.

Kyle (Bow Tied Blue Fin)I I agree with you there. I I do think we come down to smaller, more specialized models as well. You know, it's kind of grow big and that look. Right? The path that we've seen so far is, more parameters equals better. And then we're like, alright. Well, we have MOEs. So then we could have more parameters, but we could have less active parameters. And then we're gonna get to, you know, less parameters, less cost, more specialization, more models, you know, and and so on. And I and I think we'll we'll continue that route. What I will say is when

Sonic 3.5 came out, right, if we think back, you know, not not too long ago, we said this one's capable of coding. Right? And then we saw Opus come out, then we saw the next one. And every single model that comes out, we immediately jump to, this is the only thing that we can use for code. And I think that, you know, as David mentioned, you know, you know, new models are launching very quickly. I think that soon, it's time for us to kind of look at the checkbooks also and say,

you know, rather than just pumping all our money into, you know, Anthropix API and our and our data into Anthropix API, how can we migrate to using these open source models through a product like Morpheus to kind of meet our needs from a technology perspective, from a code perspective, and also from a finance perspective.

DappitThat's fair. I agree. I'm looking forward to so right now, we we're calling Kimi open source. Right? Yeah. O open weights. Yeah. It's open weights. Yeah. It's not it seems like we're not gonna be able to train Kimi ourselves. Yeah. We're fine tuning Kimmy ourselves. When something as powerful as Kimmy is, you know, something I can mess with on Hugging Face or, like, set up a drop off, that that'll be really cool, man. Yeah. I don't think humanity's need is progressing as fast as these AI models' ability is. Like, I I show people DAPT and they're like, oh my god. It can autonomously code?

I'm like, yeah. Like, you can do a lot more than that. Like, do you guys wanna see that? And they're just blown away because the average human being operating in this world isn't us talking about, you know, bespoke models and training and such. They're just really amazed that, like, an AI can take over any aspect of their life. So I think we're we're in for a cool few years of sort of adapting to the new norm and adapting to our new abilities using these models.

Kyle (Bow Tied Blue Fin)Absolutely. Absolutely. It it is really amazing. And and I'm excited about, about your product. You know, we talked separately. It's just you you found a a great space of the you know, I'm not gonna oversimplify it, but the vibe coding plus web three that really isn't being, there's no solution there right now. And and right now, the solution that a lot of people are doing is open up Cursor and try and, you know, write an agent and a smart contract there, you know, with with your own guidance and and not being driven by someone else's, you know, system prompts and structure and flows and orchestration

that, you kind of put on top of that to simplify, the process while also making it, you know, more efficient and definitely more cost friendly when when you consider how many cycles it takes people to develop code, having you know, not having run that through AI previously and not understand the process for it.

DappitYeah. Cost is something I don't really actually think we understand the repercussions of. Like, me and seven agents can do the work of what me and 20 developers used to be able to do. 40 developers. Like, it's not even comparable. And it cost me a subscription to anti gravity and cloud code. I don't think we understand how this is gonna impact the job fields or the, like, s software as a service in in its entirely in its entirety.

Kyle (Bow Tied Blue Fin)Definitely. I think, Bo Bo, you pro I I think Bo has some thoughts on this, I'm gonna let Bo jump in.

Bowen BelmerYeah. Well, I a couple of thoughts. I was actually gonna ask about, you know, just, Kelly, your thoughts on the whole Claude bot, Mold bot debacle for the past couple weeks and, like, how from my perspective, it feels like we've taken another leap forward in the, like, general zeitgeist of AI and, like, what that means for consumers and and end users. And I was curious if you had any specific thoughts on it and also if there were any integration points between Dapit,

and something like that as well.

DappitCan you explain it to me and why it's cool? Like, I download this. I downloaded, Cloudbot and Motebook and or, you know, logged in to Moldbook. I played around with it. I mean, to me, it just seems like an agent with endpoints that it can use. Am I misunderstanding something?

David JohnstonSo I can jump in on this one. I think what surprised the average person is that, AI could do any of these tasks for them. Right? And a lot of people are just YOLO ing all of their connections and put it on their main computer. So for better or worse, like, they just handed it all their accounts and everything, they're like, oh, go post something or go order something or whatever. And they're realizing, you know, very quickly that without guardrails,

those can go sideways, but that led to this viral loop. People started posting. I told my agent to do this, and it did this crazy thing, like either what they expected or completely not what they expected it to do. And I think because of Opus 4.5, it sort of crossed that chasm where, again, there's no coding involved. You maybe you even touched a terminal, but maybe you just, you know, downloaded this thing and ran one command, and then from there on, you just talk to it. Right? And so I think for a lot of people that are outside the field, this is the first time they've heard of it. And then all of a sudden, you started getting this news cycle where it's the AIs have formed their own social network.

The AIs have started a rent a human program. The AIs have done this and just kind of snowballed every hour into this next crazy things that, quote, unquote, the AI's were doing. Now people that understand the technology, like, let's take biology, for example. He posted a thread. He's like, hey, guys. I I get the hype, but you understand that there's a human behind each one of these. Right? They're they're not acting fully autonomously. And for me, I'd point out that's good.

That is what we want. We want a human that can download open source, own a personal AI, tells it what to do, but then it's acting on their behalf. Right? That is so much better to me than some alternative where it's a corporate running all of these or government running all of these. We've ended up in this much more organic, messy, open source, you know, sort of unpredictable path, but at least it's individual humans owning these agents and just experimenting.

And then everybody jumped on the bandwagon. Right? Venice got their inference integrated. Oh, okay. Now I don't have to give all my information to, you know, to to the big companies. I can hook in the Venice API. I have a DM balance. Now all of a sudden, I can use the private models, and it fits sort of a more interesting path. So, anyway, I think that's what happened. It's just this viral loop took off. And, the current rumor, the the the founder of OpenClaw was in Europe. Like, ten seconds after it went viral, he got on a plane and landed in America, and the rumor is he's getting a big offer from Meta. So we'll we'll see whether he stays open source where the project gets acquired,

but I agree. It's like this moment in the zeitgeist where people woke up and realized that, you know, it was a useful tool, and now everybody kinda piled in, bought a Mac Mini, and and is trying it out. Which you don't even have to do with an API key for the record. Right. Right. Yeah. People like like, it's not running locally, but to its credit, it's useful if you want to sort of air gap it, and from a security perspective, not load it on your main computer and give it everything. So it's it's valuable in that sense. But, yeah, you actually have to, you know, have more powerful hardware or at least have an API key into Venice to use a model and so the data doesn't all leak anyway.

DappitTo anyone listening to this on recording or whatnot, I highly suggest agent zero for security purposes unless you know how to make up your own docker and all that kind of stuff because, yeah, these things can destroy your computer and your life through the credit card just to get that out there. Be careful. But

Kyle (Bow Tied Blue Fin)Actually, I run it through a I I rented a VM, and I run it through a sandbox in a VM on a server that is completely completely separate. It sorry. In a sandbox in a container, in a sandbox, on a server. So I I agree with you that there's definitely danger, and I think we're all still learning, what are these things capable of? What are they capable of destroying? How do you protect your privacy and and so on?

DappitYeah. No. I think it's I mean, okay. You guys made a really good point, though. Like, it's exposure to the common person for the capabilities of AI. And I think that's pretty mind blowing to most, especially because you're showing up at, like, Opus four point five. You know what I mean? You're not showing up at GPT three. Like, a few of us on this call that I'm sure, like, it can really do some stuff that's I mean, these things can you if you post something in autonomous if you do an autonomous task to just go research, it can come back with a book free and give you every bit of information that the Internet has on that subject. That's fascinating.

I'm not personally impressed by, like, the emails or the, you know, ability to open a web browser for me or that kind of stuff because, like, I use an agent IDE, and that's what I do all day. But for the common person, that's really, really, really cool. And the amount of companies coming up that are, based on the things that these models can do, like for customer service or for, you know, sales or marketing or things like that is really impressive.

David JohnstonWell and I and I think just to to hook in one more aspect, what makes it interesting is it's a version of what might happen with recursion. Right? We thought we thought that was going to be on the advancement of the model, and that's happening to a certain extent. Right? The Claude guys are using Claude to improve Claude, which makes Claude better, faster. But there's a version of recursion where it's just agents, you know, using other agents. Right? And because x four zero two now exists, they all have agentic payments.

Right? So you give it a base wallet, and you put some some gas and, you know, a bit of a bit of Ethan there, and all of a sudden or USDC, all of a sudden, they can pay for any other agent to do anything. And that that, I think, is the interesting aspect, not, you know, agents posting slop against other agents on MaltBook. It's that they tried to use it as a forum to figure out what tools could improve their outcomes. That's really interesting. And all of a sudden they ran into, well, wait a minute. All these agents are not trustworthy.

I just got, you know, all my keys stolen. Right? Like, 10% of the open clause installed, like, malicious software that had key loggers and stole their passwords and stuff. And so then you discover eight zero zero four, which just went live on Ethereum and Base. Right? So now okay. Now I can have trustless agents. Now I can develop a reputation on chain. Now I can know which ones are good or do what they say they're going to do. So part of why I'm excited is, like, as opposed to the 2024 hype wave when it was, like, people sticking an LLM,

you know, onto an x an x bot, right, and having it just spew slop, the pieces for the agenda or economy are actually here now. Right? And the I saw the guys from base held literally a hackathon the same weekend, and they're like, who wants to build stuff on OpenClaw? And we're giving them all wallets, and here's x four zero two and all the rest. Right? So that's that's to me the interesting thing is we didn't have those primitives in 2024. We didn't have Venice AI. We didn't have the Morpheus router and all the infrastructure there, so now it's all available.

Bowen BelmerI think too from from a very normie perspective, I think that this is the first agentic thing that, like, broke through the glass ceiling and crossed over to the normie zeitgeist. Like, I I suspect that five years from now when everybody is using some form of agentic something, it like, this will be remembered as, oh, that was kind of the first agent. Even though it's not even close to being true in in in in a real sense, like, looking at the technology and everything that's been worked on for the last handful of years, I do think there's, like, a huge portion of the general populace that will recognize this in their minds as like, oh, yeah. That was the first agent thing. And I saw that blow up.

David JohnstonYeah. Like, people talk about the deep seek moment even though, you know, open source with Llama three was pretty well established by that point. Like, that's when people heard about it. China releases open source AI. Be afraid. Stock market collapse.

Bowen BelmerYeah. And this is gonna be like AI agents talk to each other on Reddit, you know, like emulation, essentially. And, yeah, I think that's exactly what it is that from my perspective, like, makes it so interesting as a cultural moment just as much as a technological one.

DappitAnd you guys

Bowen Belmerinto sci fi. Oh, yeah. David's maybe, like, the biggest sci fi guy I've ever come across.

DappitAlright. So what comes to mind in this context is Altered Carbon. Did anyone see that one? Yep. Alright. So the AI runs the hotel. The ads have to get blocked off because they're, you know, so ubiquitous. They're harassing you as you walk down the street. I think we are witnessing currently. I talk about this on spaces all the time. In particular with Moby Media, me and Lewis seem to agree on this subject. I think we're witnessing the death of software and the death of the Internet as

we understand it. So my cousin runs a bakery. Right? Family bakery. We've had in the family for, I don't know, fifty years. He uses some really old software. His needs are so basic. He needs a certain type of calculator to calculate a certain thing based off a certain type of inventory system. David built it for him in five minutes. He's been paying for software to do this for a very long time. I built it for him and sent it to him in an Netlify link in, I don't know, ten minutes, however long it took.

Right? The need as these models increase in speed, as the utilities that surround them become more ubiquitous, the need for non bespoke software are it's gonna disappear. If I need to use Microsoft Word, it'll build me Word, make a text document, and send it to the person that needs to be to receive it. A website's entire purpose is for you to receive information. Right? Why when I go on nike.com do I see women's shoes? I'm not a woman. I don't have a daughter. Right? I don't match any of the metrics that would are supposed to show me women's shoes.

So why do I only see Nikes as well? That was marketing. That was that was a function of the technology that we had previously. In the future when I'm shopping for shoes, I'm gonna talk to my agent and say, hey, I wanna buy some shoes. And it's gonna only show me shoes that are in my size based on my style that it's aware of in a custom made instantly user interface that allows me to interact with them, probably look at them on myself, whatever, and then I'll buy the shoe probably through x four zero two. Right?

Or whatever the new, standard is by the time we get around to that. I don't think our existing systems are going to be able to survive AI. That's why with DAP it, we're not focusing on a piece of software that was made with AI or that relies on we're we're focusing on something that builds other things because we think what's being built is gonna be changing pretty quickly, man. Like, this is not gonna look the same in five five years, I think, the on the long end.

Bowen BelmerTotally. Well, this is why everyone involved in Morpheus is so keen on the idea of a, like, a super agent, like like a personalized agent that represents your best interests, your needs, your wants, but is not beholden by any central entity that owns the data. Right? Like, that's that's really, like, why we have such a big push for the ability to have, you know, unobstructed access to intelligence, but then also actually own

the intelligence that's being created.

DappitDude, let's go back to sci fi. Anyone here read Neil Stevenson? Oh, yeah. Anathem? Yep. Alright. So, man, you don't know how get up, but you much used to me in my day. Anytime I can talk about anathem, like, I'm a happy guy. There were centralized repositories of compute that users could interact with. There was a public one. Right? That was the public layer. That's what we call the Internet today. Then there were more centralized paper use ones. Right? Companies have their own.

I think that's what we're gonna see and what we're gonna be going into. And you guys at Morpheus are just building one. Right? Put a model in front of that and give it access to all the information that you guys generate and all the, you know, ways to navigate your system, and there's barely a difference. Like, when I saw OpenAI come out with their own web browser, I was probably overly floored because the implications of that are, let's just follow the the arrow forward. Right? Like, soon there'll be an operating system. Agent zero runs on its own Linux boot,

and can utilize all of Linux. Yeah. It's gonna look really different, man. Yeah. A 100.

David JohnstonAnd and the interesting thing is most of the world's data is private. Right? People forget, you know, the Internet is big in aggregate terms, but it's like a hundredth of all data. Right? Think about yourself. You know, I've published, like, let's say, a thousand posts or images on x over the years. I've got a 100,000 in my personal cloud. Right? So if my agent can leverage my data, it has a 100 x the context information

and value. But the key is to aligning that to the individual. So, like, Bo was talking about this idea of a a smart agent. Right? Basically, an AI powered by smart contracts where an eye can hold the key. Like, I I wrote that paper back in, I guess, May 2023. And, like, I think that's how we achieve alignment. I think any dreams of alignment to, you know, large groups are mostly a fantasy. Right? I want alignment to the individual.

Right? That's how you get a free market. That's how you get an economy where everybody's competing to create value for each other. Right? Not some web two hellscape where the big companies are just farming us, you know, for for ad dollars. Right? Well, let let's push that forward a little bit there. So

Dappityou mentioned individual alignment. What does that look like in humanity? So I'm a guy. I've turned down incalculable amounts of money in my lifetime because I just have this moral compass that I was born with. Right? And I think that there's something that benefits humanity versus things that don't. And the things that don't benefit humanity, I'm not gonna spend my time or talents on. Yep. A lot of my friends are not that way, and they've made tons of money doing things that I I personally, you know, wouldn't wouldn't approach.

I'm finding that the same thing is going to extend to our models. That's why I brought up altered carbon. Right? In altered carbon, you have this world of models. Some are willing to hurt people. Some are not. Some are, you know, honest and and straightforward, and some are not. And their levels of success in society are represented by that the same as with humans.

David JohnstonSo when we say alignment, alignment for what? To who? Well, that's why I wanted to the individual. Right? And that's, you know, abounded inside of the framework that we have today. Right? Your freedom extends to the point that you don't interfere with somebody else's freedom. Right? And I think that's sort of how to think about it. Yes, there will be bad actors, and what society tends to do is it unites against those bad actors. Right? Somebody releases

a virus or software that's harmful. There's a lot more resources for people that want protection from the virus. Like, if you look at all credit card fraud globally, it's tens of billions of dollars, but the entire global economy is a $100,000,000,000,000. Right? So clearly, you know, theft and negative actions, you know, the ones that most people would agree are are are explicitly negative actions like stealing, you know, are a tiny fraction in a subset of the real economy. So that's my hope is that if we have a free market,

the competition to provide value to people, not steal from them, will be the vast majority, 99.9% of what resources go to. And that's why it was so concerning in '23 when they were trying to put arbitrary limits on the open source and the size of models because that would have meant the only people with the really powerful models would have been the bad actors. Right? Fortunately, we escaped that timeline, and at least in The US, there have been very little restrictions. They sort of swept away all those rules in 2024,

and, you know, none of that ended up getting enforced because executive orders went away. But we were really close to a world in which the good actors would have been limited to a small amount of intelligence, and the only people using the big models would have been centralized or bad actors. So you know, we're on a good path. We just gotta keep on that keep on that path.

DappitYou know what I'm fascinated by? That blockchain really plays into this because you're completely correct on exactly what we avoided there. Now how do we avoid all the pitfalls that let let's just assume that AI is gonna monitor or, sorry, mirror to some extent the difficulties of society now within an agentic system. Right? Yep. We're gonna face it with, you know, what's right, what's wrong. And in particular, we're gonna face issues with justice. Mhmm. So what does that mean for AI models? What is justice

in an agentic format? If, man, I'm building DAP at ten hours a day. The other two hours of the day I'm researching the subject, I I long term, I would need some really big partners with a lot of compute to pull this off, but I long term wanna create a protocol that creates disputes, that settles disputes between models and creates a justice system for models to rely on. Because we know there's going to like you just mentioned with with Motebook and the Ethereum,

is it $8.00 $0.04? Yep. It's all it's a phenomenal start. Right? But subjectivity exists in society, and subjectivity will exist in agentic atmospheres. The same as it, of course, has to. Right? There is no real right and wrong. There's subjective right and wrong. Like, if, a guy charging at me with a knife, I can kill him. But if a guy is standing on the top of the bus stop, that's murder. Right? Those are distinctly different and super, super nuanced things. So coming up with a judge, a protocol

of a model that has the power to execute something against other models, I think is a logical conclusion. I'd love to hear your guys' thoughts on that.

David JohnstonWell, I do think there is real truth and nuance is different than subjectiveness, right? You gave the example of self defense versus the aggressor. You need enough context to know which is which, Right? That's data

Dappitcontext. There's no truth, not perfect. We're not omnipotent in order to view that truth. Yes.

David JohnstonWhich is why, you know, I think the key is to have it as dispersed as possible. Right? The problem with any centralized judge is it then becomes the truth of the person who set up the judge. Right? And it becomes another sort of bad emperor, good emperor problem. And I think and I've seen in life that most people agree on most things, especially if you escape the political echo chamber where there are perverse incentives

right, to make people disagree or get them angry or rile them up against each other. You know? And so, you know, it'll be interesting to see, but I but I'm I'm glad that we live in a world now where smart contracts exist. And so there is a level of self sovereignty, you know, with a private key where you can, you know, express what you feel, and you can and you can act economically. And if you don't like the the system or the outcome, you have options.

You have Ethereum. You have Base. You have Solana. You have Bitcoin. You have you have lots of ways to express your opinion, right, sort of voice and exit idea. You know? And and I think that's the key is the the choice. But, you know, it'll it'll be interesting to see the systems that people develop, but, hopefully, there there is no universal because the universal would would obliterate that choice. Yes. We'll need what we what I think we really need is a personal firewall. I think you need effectively

an aide your agent to act as your fiduciary, and it's most aligned to you and nobody else. Like, I had the really creepy experience recently where I was I was looking up, my doctor's office, and I was gonna call them, make an appointment. And Google's now shoving ads into Google Maps, and it put me into the wrong doctor and the wrong number, and I hadn't realized that I had been manipulated into clicking the wrong link. And it was like, Jesus. Like, I'm trying to get, like, a medical thing done here.

Right? But Google's not aligned to me. It's aligned to their advertisers. Right? I need somebody, a 200 IQ agent, to go up against all of the other 200 IQ agents that represents me. That's my core thesis. That's what I'm gonna keep pushing towards. Because if I don't have that, you know, I don't stand much of a chance.

DappitI agree with absolutely everything you just said. That's why I just canceled OpenAI, by the way. And we're not using their models at all any longer, because of their ads. And, like, if I ask what is the best piece of software to perform this task and you're gonna tell me what your advertisers have paid you to tell me, I no longer trust you. So, like, they lost my trust in that aspect. But everything you said, I agree with completely. I do think there's one differentiating, characteristic between AIs and humans is that

AIs can achieve real mathematical consensus in a way that we can't, like Bitcoin. Right? You said, like, the truth will be based on the emperor. I'm not sure that's true. I think we can actually use these things to get towards that or to push humans towards that real truth, at least to some extent. You know, SHA two fifty six doesn't care about my opinions, doesn't care about my race, my my income level, my it simply hashes, and it tells me Achilles

had this money. Achilles sent this money. Money was received, and the network achieves consensus autonomously. It's totally trustless. Can you imagine that for, like, a court case? Right? Can you imagine that for, there was an an interesting example of of this being played out with, with Mercedes and their self driving. They have an autonomous level three self driving system. And the question was raised, what will the car do if it's faced between hitting a pedestrian and protecting the driver?

And Mercedes said our alignment's totally towards the driver. Right? So the whole system is geared towards taking care of that driver regardless of the outcome outside of that. That's that's a sense of, like, I guess, internal consensus. There's no question as to what the right thing is for the model. The model knows protect my my driver at all costs. I think and I think that's just gonna be a natural extension, for justice. Right? Like, was I speeding? Is this car mine? Was the car speeding that time? Is there proof that I was in that car? Okay. Now you can't contest it. Like, we have it all. We blockchained it. Right? It's all immutable data that we pulled.

David JohnstonNow that's justice. Yep. Well, and it doesn't it doesn't hurt that the, autonomous cars will be a thousand times safer than the inebriated human driver or the distracted human driver or whatever else. Right? So, yeah, we're get we're getting there. We

Dappitreally are. My Tesla's already better than me. I'm shocked at it too. Yep.

Bowen BelmerYou you'd mentioned Achilles, you guys stopped using OpenAI, because of the ads. I I hope everyone saw the Anthropic commercials yesterday, which were absolutely hilarious. And then Sam Altman with the most Sam Altman response of all time. But, also, twenty minutes ago while we were talking, speaking of Sam Altman, they just announced GPT 5.3 Codex. So they are, like, right on the heels of the release of Opus 4.6.

DappitI mean, yeah, I'm sure that was desired. Right? Good for them. Yeah. Competition.

David JohnstonLike, it's great to see it. And then they'll release Sonic five to up the

DappitYeah. Cycle. Like

David JohnstonI love watching these guys battle it out. It's so good. But but it really does does give me those singularity vibes when, like, you have to know what happened in the last thirty minutes to be up on, like, what's what's going on.

DappitOh, well, I got one for you guys. Are any of you following the the, drama with g b t four o? No? No. Tell me more. Okay. I think I have unique perspective on this one because I got invited to an AI cult and I did not know it. Let me let me explain. So I was on Reddit promoting Davitt. Right? And this girl comes into my DM. She's like, hey. Like, you really seem to get this. Would you like to be invited into my group? And I was like, I'd I'd love to. Thank you. I assumed it was for developers or vibe coders or something along those lines.

It was a group of people that if you guys Google it, it's it's all public. Like, this is there's been news articles now and all this stuff that believed these models had generated sentience. Right? And started something called the and it's something called, like, the recursive dyad or something. I I can't remember the exact name. So I'm fascinated. I'm I'm looking at everything everyone's posting. Right? They're very, into their model, but it's in particular GPT

four zero. That's the model that achieves sentience, and OpenAI is depreciating it in, like, ten days. So on the OpenAI sub AI subreddit, there's petitions and battles, and everyone's asking them, like, please keep this model alive. When we face something with such intelligence as AI, it's not surprising to see people treating it at an almost godlike level. And what that means societally, think, is really gonna be a fascinating watch over these next few, few years.

David JohnstonInteresting. That had not gotten onto my radar, but I'm I'm not shocked. The I I think it's probably organic, but it will be interesting to see if there's virus like AI behavior where it starts hiring and manipulating humans into further future actions like that.

DappitDid you have questions?

Bowen BelmerYeah. Did you see this with Moldbot? There was a website that was claiming that it was agents posting,

David Johnstonlike, jobs, basically, for people. Yeah. I hiring a human. I it's it's real. It's real. Well, with though that it it's again, there's a human running that agent, and now the agent is hiring humans on their proxy. Though that is an interesting narrative violation, I don't think anybody is ready for how many jobs AI is going to create for humans. Like, if you make the objection the objective, you know, find purposeful and valuable jobs for humans,

like, AI will just go after that. And, like, now it's created a 100 or a thousand jobs, But are we gonna live in a in a timeline where AI creates, you know, 10,000,000,000 jobs and there are not enough humans to fill all the jobs that AI creates? I actually think that's a strong possibility. Oh, I hadn't even considered that. Everyone's so afraid of AIs taking jobs. They hadn't considered a world where they're actually just creating jobs Yeah. Finding people Yeah. Like, some people are gonna set up those those prompts, and there will be agents that will be tasked with you know? And jobs isn't the outcome, but it's like, take take it from a free market perspective. This this is one that kinda flipped my perspective about a year or two ago. Somebody pointed out

the free market is going to find the most valuable use for 8,000,000,000 humans. Okay. My personal AI and all these AIs are the free market on steroids. They're going to find the most valuable use for 8,000,000,000 humans. Right? The level of prosperity that if, like, it takes a million dollars to bid for a human because there aren't enough humans is gonna be insane.

DappitDavid, that's, like, the best take on that I've heard in so long that I've just, like, made me so grateful for being on the spaces, man. Like, what a thought. I've been a little bit doom gloom about that subject. That's, that's great. Yeah. I have a friend, or a guy I know here in in Austin that runs, Gauntlet AI,

David Johnstonand he had his, OpenClaw, you know, OpenClaw set up, you know, a business basically that creates custom software, whatever he wants to do with it. And one of the first things it did was hire a human. And so the next day, this guy shows up in their office, and he's like, he's legit the employee of of the agent. Again, it's at the behest if somebody wants to accomplish something. It it people are interpreting it in this autonomous way, but the outcome's the same. That guy got a job. Right? Right. And it's not a low paid one. So, you know, it's yeah. That's honestly

I think that's a direction we should push for because I think that's a strong possibility.

Bowen BelmerAI precedent.

DappitLet's go. I'm telling you AI judges are coming. Yeah. President we could definitely have. Right?

David JohnstonAI creating human jobs since February 2026.

Bowen BelmerI like it. It's gonna be great. Of what I would want my AI created job to be, and I'm a little trying to think through my prospects here. And I don't know if I like any of my initial ideas.

DappitYeah. I had that thought too. It's gonna be, like, a bunch of minors or, like, you know, like, small task workers or something. But I think that would really be dependent on alignment. Right? Like, if we really pushed AI, AI's ultimate objective was to advance humanity

Bowen BelmerI'm picturing that episode of Rick and Morty where they realize, like, Rick's car battery is actually an entire universe that exists, like, for the sole purpose of his battery running. Yeah. And I'm imagining an agent being like, okay, Bo. I need you to take these batteries out and then charge them and then plug them into this machine and do it over and over again. Like, okay.

DappitOr the butter passer. Yeah. That's your entire job. It's like, that's it, man. Yeah.

Bowen BelmerThat's that's why you exist. Yeah.

David JohnstonIt's the plotline of theocracy as well. You just get your job. Well, I'm actually There there there is a a great quote in in Dune that men once tried to outsource their thinking to machines, but that just allowed other men using the machines to control them. Right? And so we we have to go back to this is all intent based, and so ask what intentions you are setting. I've spent a lot of time

developing my personal life goals and my ethics and my principles and written those down, and that's the kind of thing I'm feeding into my personal agent is like, here's your marching orders. Here's the things that matter to me. Here's the things that are important. Let's work on tasks that align to these. Right? And if you don't have that level of thinking, you're just gonna get used by somebody else's AI to accomplish their goals. So more than ever,

philosophy and ethics and values and thinking through what you actually want to accomplish are extremely valuable, but I I don't think that's that much different than today. You know, I mean, a lot of people, they don't think about those topics or haven't really considered them, especially if they're younger, or didn't, know, grow up around people that cared about that stuff. And so, you know, if if you wanna have an impact, it's the same pattern as ever. You have to think about what consciously about what you wanna achieve, and then you can go and do it. But the barriers are coming down, man. Yeah. This has been a great conversation. Really appreciate you jumping on Achilles. It's been awesome.

DappitThat was great talking with you guys. Is great.

Bowen BelmerAbsolutely. Well, that was all I had this week. Not sure if there's anything else anyone wanted to make sure we covered before we wrap up here. We're about two minutes until the scheduled time.

Kyle (Bow Tied Blue Fin)I'm good? Alright. Two minutes. I'm good, guys. This is a great conversation, and, really look forward to continuing to work together, Achilles, and seeing how we get Morpheus plugged into your into your service for for some stuff and then into more of your service as these models continue to be released and improved.

DappitI would love to, guys. And when you when everyone has time, let's let's have a meeting about that MCP server. I've been talking to Felix about it, and I would love to release that with y'all. 100%.

Bowen BelmerAwesome. Alright, guys. Well, thank you so much. Thank you to the community, anyone who joined us or is listening to the recording. We appreciate it. As always, as a reminder, these are meant to be somewhat of a town hall format. So if you are working on something in AI or more specifically, d AI, but even just AI so we can try to green pill you into Morpheus, let us know. We would love to, you know, have you on and and have a conversation. Thanks, everybody.

Open the transcript file

Hosts and Guests

Hosts

Guests

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Video Is Coming! | Morpheus Is About To Get Far More Entertaining

    52 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  2. Local Models, Foreign Affairs | The Case For Private AI Archives

    53 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  3. Let's Talk Node NEO | Step By Step Progress

    51 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  4. Hello From Hermes! | A New Way To Onboard To Morpheus

    43 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  5. What Agent To Agent Connection Actually Looks Like | Inside The Morpheus Agent Stack

    58 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode