The Rise of AI Agents. Why Web3 Infrastructure Is Nowhere Near Ready | Alice Shikova | ETHWarsaw [4]

Hi everyone, my name is Alice. I'm the marketing lead at Space ID, the leading digital denting and web predominant platform. We have more than uh 2.7 million users and 6.7 million domain holders.

I'm here to talk about something that is exploding across the crypto space and quietly breaking it at the same time. AI agents and what's holding it back. A bit about me. I'm the marketing lead at Space ID. I'm X CMO at Folks Finance.

For the past five years, I've been deeply involved into fintech, specializing in strategic marketing and product marketing for crypto and D5 projects. I co-founded a web through marketing community in Lisbon and became a mentor at women in blockchain Africa program. To set the stage, AI agents is no longer a futuristic concept. They already beginning to play a role in managing our financial assets as well as automating our daily tasks. dreadf giants like Visa adapted to the shift and they recently launched intelligence commerce um that that integrates AI agents directly into the payment flows eliminating the need for manual approval at every stop.

PayPal recently released their MCP server that allows AI agents flows into payment subscription invoices processes and other payment related uh things. But web3 lags behind and web3 um infrastructure is still taking shape to support agentic AI and autonomous software that can act on users behalf. Web3 needs a more robust infrastructure in place. Um there is no agent compatible identity wallet infra is lacking and execution layers uh is broken and the stakes are really huge. By the end of 2030, the next predicted more than 10 billion in annual revenue for crypto AI.

And as Mark Zuckerberg said, there could possibly be more AI agents than humans. So why we all here discussing this? Uh because the potential of AI and blockchain is a gigantic being totally different from one another. These two technologies already stand out as the most groundbreaking technologies of this decade. What smart contract started, AI is going to finish with flare and the convergence of web 3 and AI provides a much more robust digital economy than it could shape in web 2.

For example, in web 2 and threadfi, uh AI agents are not able to hop from one bank to another full of bureaucracy and paperwork that slow things down. And this is totally different in web 3. So why web 3 and AI is the ultimate combo if we build the right infrastructure? Um democratic access and control when the power belongs to people not the corporate boardrooms and not just Silicon Valley executives decide who get access to AI. We have enhanced privacy and data sovereignty processing happens locally without feeding corporate surveillance machine.

And I can share my personal story here. Um, as many of you, I work closely with Chad GPT and Claude and they already not just assistants anymore. They became a part of how I brainstorm, work, create, automate. And every time I would authorize Chad GPT to grant access for my files, to process my um, work documents, I would be so hesitant because I don't know where this info goes. It feels like a blackbox model I don't control on a server I don't own and I share my knowledge, my voice, my work, my creativity.

Um, and yeah, this is something that AI um plus web 3 can solve and I'm going to mention a couple of projects that are working in that direction. Um, innovation and diversity developers worldwide can build freely without corporate approval. Um as some months some months ago we remember that open AI uh put some restrictions on who can get access to um API and this should be different in web 3. economic benefits when value flows to creators not ga gatekeepers and contributors get rewarded for example for providing training data uh compute uh and other essential things uh for AI agent functionality and this is why it this is more resilient and robust. So looking at the decentralized AI landscape if DeFi was moving so fast decentralized AI is moving at 100x speed.

Today we have more than 17,000 AI tokens launched on virtuals. We have more than 500 AI related products listed on Misari and more than 1,000 tokens listed on Coin Gecko. Um representing the core frameworks like Bitensor, Singularity AI, uh Fetch AI, Eliza and the more mim and speculative games. Uh so closing the infrastructure gaps um in its report Misari bluntly outlines the missing infrastructure pieces that AI and crypto niche currently needs. Um if you look at the Misari dashboard you will see multiple niches being quickly filling up with newly with new innovative products in compute indexing uh automation and um other niches.

And how if you see uh AI agents projects they're taking the biggest uh SL the biggest um slice of the pie. Uh according to Misari over 1.39 billion was raised by AI projects in 2025 only. But here is the catch. We still haven't seen AI agents in Web 3 to actually scale and to become useful.

Um the most successful use cases for AI agent projects have been speculation and memes and there is a reason for that. The reason is that the web pre infer is not ready and this is what I did um based on the misari report. As a marketer I love building funnels. So I did the same for agents. Um I called it agent funnel or agent staircase you name it.

So it covers the base uh generally infer layer agent specific core and agent intelligence uh layer and the lower you go the more agent critical and essential it becomes we're going to cover every field and uh to get started with decentralized comput and storage AI agents need the standardized comput and storage that could scale for high performance workloads filecoin and eighth year are working in that direction but it's still work in progress eighth here is providing um AI agents with the ability. So, AIR is building a decentralized network where everyone can contribute spare GPU capacity to a global pool and then it can be later rented out by AI developers. Um Filecoin is building a decentralized network to store agents data uh model training data and other essentials. um privacy preserving computation. Um so confidential AI is something that we can reach um with the convergence of web 3 and AI.

uh referring to my past story about authorizing chipdml provides a really promising foundation for privacy preserving computation when it can prove that the computation result was correct without exposing user data um without exposing the model the output and the computation process itself. Flock IO, a decentralized uh privacy preserving federated learning platform that allows to train models without exposing user data. Um and we definitely need more projects like that. Interoperable ecosystems that are not um present for now. We always have interoperability issue.

We still have it. Uh I hope we will not have it in the future. uh and we still don't have established standards and interoperability protocols that would allow agents to operate across different platforms and blockchains. Um however um one example is chain link CCIP crosschain messaging uh interoperability protocol that was already used by Eliza OS agents um and it enabled them crosschain functionality uh across Salana ecosystem and beyond. incentive and governance.

AI agents ecosystem need um robust crypto incentive frameworks to sustain themselves. We need to incentivize operators to host manage AI agents including permissions uh dispute resolution and uh other things abstraction and usability trust identity and provenence autonomous execution. It's more agent specific core and I'm going to uh spell it out in more detail. abstraction usability. Um so we still haven't built the infrastructure that would abstract uh the complexity of underlying um crypto and blockchain infrastructure and deps platforms and wallets were not built for agents.

They were built for humans. For example, wallets are not scalable for AI agents. They need um the approval of us humans almost at every step. Um, and what agents need instead, they don't need a MetaMask popup every 30 seconds. They're going to they're not going to work in this way.

They need programmable approval rules. Uh, for example, pre-approved swaps under 500 or delegate all our voting to this logic tree. They need multi- aent permission systems when multiple agents have coordinated access to the wallet and for example one agent triggers a signal and then another agent executes and they need scheduled event trigger action. Identity and reputation system. This is the core missing layer for agent trust.

AI agents need searchable verifiable and unique identities. For now there is no shared identity registry. Agents lack a universal naming system. Discovery and execution is broken. There is no place to discover agents.

There is no universal marketplace. And this is how the discovery in web3 AI looks like today. AIXBT agent launched on virtuals with a hull string of wallet address. That is not readable. That is uh super easy to mistake, mistype and uh actually lose your funds.

So we got to change that and it's disconnected from reputation because there is no identity registry because uh agents don't have naming systems. We cannot track agents history performance and build trust score and without structured agent identity we get blind delegation to unknown agents and impossible multi- aent coordination. And regarding multi- aent uh coordination uh this is the example of uh how multiple agents coordinate and operate with each other. For example, we have agent A who is a portfolio manager and it looks for the best yield farming opportunity. Agent B that is a DeFi scanner claims 15% API on new protocol and agent C risk assessor warns about smart contract risks.

For those three to operate seamlessly uh and in a reliable manner uh we need to have identities. We need to have verification scores at the stations. So this is the example of how day3 can filter each other by reputation, performance history and credentials. And on the right you can see the example of a verified agent with a trust score, profitable trades, transactions and asset under management. So because of the things that I mentioned earlier, the execution is fragmented and there is AI that can understand us.

There is blockchain that can execute but there is no middle connecting the human or agent intent with the actual onchain action and that's the NLP to transaction gap. Um so for example when I say swap 50% of my USDC to ETH if ETH drops below 3K right now that's fragmented across five different tools chat agent price oracle smart contract wallet UX interface and nothing bridges the full flow but there is something else uh that could bridge the full flow a catalyst that could turn the isolated layers uh into a system where agents can collaborate build improve across all the fronts and that's the missing there um uni unified data so the data context is fragmented everywhere and blockchain blockchain is transparent but it's not readable and AI agents need high signal structured readable data um for now there is no unified data layer and for example from this query the agent doesn't know was it a unis swap swap was triggered by liquidation? Has a user just bridge funds from optimism? Was it frontr run by a bot? Um, and raw blockchain fees don't work.

Plus, we need diversity in data. We don't need just blockchain data. We need social data, off-chain data, crypto, stocks, all of those data types would allow agents to act autonomously and to make really intelligent decisions. data is also scattered across different APIs and it comes in different formats in inconsistent schemas. Uh here you can see different um examples of API from Binance, Coin, Gekcko, Polygon, Twitter.

uh it all comes in different formats and it's it's a waste of time for AI agents that instead of reasoning they have to spend 80% of that of their time on normalizing different schemas and custom custom integrating every new data source. They cannot reason across different markets if they miss um a type of data for example uh social data and they can miss critical context and make poor decisions. Uh here you can see um an example of how important data is uh to enable full capabilities of LLM and AI agents. AI's performance is directly tied to the quality, breadth and timeless of its data. How can AI agent complete a proper analysis if it doesn't have all the data it needs to create the full picture?

I asked Chad GPT if it has access to stock and crypto market data to build an investment and trading strategy for me. Um what Chad GPD answered was that it doesn't have live access to market data and Claude said that um let's see what it says. I don't have direct access to real-time stock and cryptocurrency market data through dedicated financial data feeds. However, I can build I can help build you the strategy. So, this is definitely not what we uh want from AI agents.

We want them to do things for us and uh to act autonomously. And for this, we need to feed them with normalized structured high signal data to take them to the next level. So this is how we came to building arrays um at space ID a universal a unified data layer that delivers uh high quality high signal real-time data for AI agents. A bit of a context space ID is the biggest web3 domain and provider and digital dentist digital dentistry platform. It has been a pioneer in digital dentist space since 2022.

We have more than 2.7 million users and digital identity was always at our core product offering and it allows us to understand that identity is a part of data. But to build identity for AI agents, we need to have an established market that doesn't exist yet in web 3. And we decided to give the agents real utility, the real power, the real engine that would take them to the next level. And we build a race So arrays would bring the full market context to AI agents covering market data, company data, financial data, news, uh info, social.

Uh we are adding more and more data API um into our stock. One API will provide the whole context AI agents will need to act uh autonomously and to make intelligence decisions. It also have um multi-asset coverage and real time plus historical context layer. So yeah, for now um we we're looking at this as a human readable agent ID and unified structured data put into MCP and API for AI agents to function seamlessly and without friction across web 3 and beyond. If you've noticed today um many projects have tapped into AI niche.

Uh even OG companies everyone tested AI either launching new products or expanding their product offering and that's totally fair. AI is here to stay. It's not going to go away. It's our present reality. But we need to understand that we need to build infrastructure not for humans but for the next generation of users machines.

Thank you for your attention.