X Space with UniLend Finance — ChainAware co-founder Martin in conversation with Ayush from UniLend Finance on revolutionizing Web3 with AI agents. Listen to the full recording on X

Two AI agent builders from different corners of the DeFi ecosystem sit down to map where Web3 is going. Ayush from UniLend Finance brings four years of operating a permissionless lending protocol and a new platform — LLAMA — designed to let anyone launch AI agents on blockchain without writing a single line of ML code. Martin from ChainAware brings the perspective of a team that built AI agents organically, block by block, starting from credit scoring and arriving at autonomous marketing and compliance agents without ever having “become an AI agent company” as a stated goal. Together, they work through the questions that matter most: why 95% of token holders never touch DeFi, what makes Web3 structurally superior to Web2 for AI agent deployment, how the convergence of real-time data and autonomous operation is creating an economic shift comparable to the internet itself, and why the innovation wave that is just beginning will emerge from Web3 — not from the closed systems of Web2.

UniLend Finance: Four Years of Permissionless DeFi and the LLAMA Agent Platform

Ayush opens the conversation with an overview of UniLend Finance that immediately establishes the platform’s credentials: a DeFi protocol live on blockchain since 2021 — one of the longer continuous operating histories in the DeFi space — with approximately $4.2 million in Total Value Locked on its V1 product and a recently launched V2 that introduces fully permissionless lending and borrowing.

The V2 product takes the permissionless model to its logical conclusion: any token can be listed and used for lending and borrowing instantly, exactly as any token can be listed on Uniswap for trading. No governance approval. No whitelist. No manual curation process. Just as Uniswap’s permissionless model democratised token trading, UniLend’s V2 aims to democratise yield generation — removing the gatekeeping that has historically kept most DeFi lending products accessible only to tokens that cleared a listing committee. Beyond the core lending protocol, UniLend is preparing to launch LLAMA: a platform that enables anyone to build and launch their own AI agents on blockchain without prior machine learning experience or agent development skills. As Ayush describes it: “You can build your own AI agents and you can launch them directly on blockchain without any experience in developing agents or learning ML. You can just directly go and launch your agents.” For the full context of permissionless DeFi and how AI agents fit into it, see our DeFAI guide.

LLAMA: Task-Oriented Agents, Not Just LLM Wrappers

Ayush makes a pointed distinction about LLAMA’s design philosophy that separates it from most of the AI agent platforms flooding the Web3 market. Many existing agent platforms are, in his assessment, effectively LLM interfaces with a Web3 skin — they can produce text, answer questions, and converse fluently, but they cannot reliably execute tasks. LLAMA’s focus is specifically on task-oriented agents: agents that complete defined objectives, trigger on specified conditions, and produce measurable outcomes rather than conversational outputs. As Ayush explains: “A lot of agents are just kind of LLMs only — they will do the talking. They are not very task oriented. So that is our focus on LLAMA — that these agents will start to help the users, meaning that people will start to work with much more high-qualitative tasks instead of doing all this repetitive data analysis.” For how task-oriented agents differ from generative AI wrappers, see our attention AI vs real utility AI guide.

ChainAware’s Journey: From Credit Scoring to Web3 AI Agents — Block by Block

Martin provides the context for how ChainAware arrived at its current position as a Web3 AI agent provider — a journey that, like UniLend’s, was driven by solving real problems rather than by targeting a narrative. The origin, as always, is SmartCredit: the DeFi fixed-term lending protocol where the co-founders first needed credit scoring models to assess borrower reliability on-chain.

Credit scoring required fraud detection as a foundation — you cannot score creditworthiness reliably if your fraud detection is weak. Building fraud detection revealed that the same predictive AI architecture applied to pool contracts could predict rug pulls before they happened. Rug pull detection revealed that the behavioral pattern recognition could extend to user intentions — predicting who would borrow, lend, trade, or stake next. Connecting those predictions to a content generation layer produced the marketing agent. Applying the same continuous monitoring capability to compliance produced the transaction monitoring agent. As Martin summarises: “ChainAware started from credit scoring, then the fraud, then the rug pull, then user behavior prediction — always building new components, always innovating, the same as UniLend. Continuous innovation. And now we are here doing the Web3 agents.” For the full platform architecture, see our ChainAware product guide.

The Platform That Emerged Block by Block

ChainAware Prediction MCP — 18M+ Personas, 8 Blockchains, 32 Open-Source Agents

Every product that emerged from ChainAware’s organic discovery process — fraud detection (98%), rug pull prediction, wallet behavioral profiling, marketing agents, transaction monitoring — accessible via a single Prediction MCP. Natural language queries. Real-time responses. 32 MIT-licensed open-source agents on GitHub. Any developer or AI agent integrates in minutes.

Get MCP Access View 32 Agents on GitHub

The 95% Problem: Why Most Token Holders Never Touch DeFi

Ayush frames the core problem that AI agents in Web3 must solve through a striking observation about the gap between crypto participation and DeFi participation. Consider a representative audience at any Web3 event: virtually everyone holds cryptocurrency in a wallet. Now ask how many of those same people actively use lending, borrowing, or yield optimisation products. The number drops by roughly 95%. Despite holding assets that could be generating yield continuously, the overwhelming majority of crypto holders simply do not engage with DeFi protocols. As Ayush observes: “How many people are actually using any lending and borrowing service? I think almost there is a huge drop — almost like 90, 95 of people who are holding any tokens are not lending or utilising any yield optimising products. Only a handful of OG DeFi users are doing that.”

The reason is not ignorance of the opportunity. Many token holders are aware that yield farming exists, that lending protocols offer interest income, and that their idle assets could be working harder. The barrier is practical complexity: navigating multiple chains, evaluating which protocols are safe, understanding liquidation risks, managing gas fees, and staying current with rapidly changing rates across dozens of protocols. Each of these steps requires specific knowledge that most users either lack or find too time-consuming to acquire. Consequently, the DeFi opportunity remains concentrated among a small cohort of technically proficient early adopters while the majority of potential participants stay on centralised exchanges earning nothing — or worse, holding assets in wallets that generate zero yield. For the full context of DeFi onboarding challenges, see our DeFi onboarding guide.

AI Agents as the DeFi Accessibility Layer

Ayush’s argument is that AI agents are the specific technology that can collapse this complexity barrier. Rather than requiring users to learn protocol navigation, cross-chain bridging, liquidation mechanics, and rate comparison, an AI agent handles all of these functions autonomously. The user specifies a goal — find the best yield for my USDC across all available protocols — and the agent executes the entire process: identifying options, evaluating security, selecting the optimal protocol, executing the transaction, and monitoring the position. As Ayush explains: “A lot of user-related problems where finding a good yield optimizing product and figuring out how secure it is and figuring out which chain you want to lend and which tokens is more beneficial — all of these things can be easily passed on to AI agents rather than us figuring out and juggling between different DeFi protocols.” For how ChainAware’s fraud detection integrates into this agent stack, see our fraud detection guide.

AI Agents Are Not a Hot Narrative — They Are a Natural Development

Both Martin and Ayush converge on a perspective that distinguishes their analysis from the typical crypto hype cycle framing: AI agents in Web3 are not a trend that smart projects are jumping onto because the narrative is hot. They are the next stage in a technological evolution that has been unfolding step by step, each stage enabled by the infrastructure built in the previous one.

Martin makes this argument with specific reference to ChainAware’s development trajectory. The team built agents not because they set out to be an AI agent company, but because each product component they built — predictive models, behavioral profiling, content generation, continuous monitoring — naturally combined into an architecture that turned out to be what the industry calls an AI agent. As Martin explains: “It’s not about that we are jumping on a hot topic. It’s about that we are talking about what we are building, what we have built.” Similarly, Ayush frames the agent emergence as a technological inevitability: “This is like a natural, you can say, natural development that is happening. There will be a lot of agents, the applications will be full of agents.” For the complete ChainAware agent architecture, see our AI agents roadmap.

From Prompt Engineering to Autonomous Agents: What Actually Changed

Martin provides a precise technical history of how the AI landscape evolved from the prompt engineering era to the autonomous agent era — a history that explains both why agents are emerging now and why they were not possible two years earlier.

The LLM era, beginning around 2022-2023, introduced the concept of interacting with AI through natural language prompts. This was genuinely transformative — but it had a fundamental operational limitation. Every prompt required a human to initiate it. Prompt engineers became highly paid specialists who could craft inputs that extracted useful outputs from LLMs. The underlying models, however, operated on training data that was 18-24 months old — meaning the AI’s knowledge of the world was perpetually stale by the time any user accessed it. Furthermore, the process was inherently sequential: human writes prompt, AI responds, human evaluates, human writes next prompt. This made LLMs powerful tools but not autonomous agents. As Martin explains: “There were people paying huge salaries to prompt engineers because it was so new. But you need always a prompt engineer. And the LLMs were 18-24 months delayed in their data.” For the complete generative vs predictive AI analysis applied to Web3, see our generative vs predictive AI guide.

Three Changes That Made Autonomous Agents Possible

The transition from prompt engineering to autonomous agents required three specific changes to occur simultaneously. First, data latency had to drop from 18-24 months to real-time — agents operating on stale data cannot make useful decisions about current DeFi rates, current fraud risks, or current market conditions. Second, the operational model had to shift from human-initiated to continuously running — agents that only operate when someone submits a prompt are still fundamentally human-dependent. Third, feedback loops had to be integrated — agents that cannot learn from whether their outputs produced the desired outcome will not improve and will not maintain relevance as conditions change. All three of these changes occurred across 2023-2024, creating the conditions for genuine autonomous agents. As Martin describes: “We have now real-time data. And then instead of using the prompt engineers, you do it continuously — you don’t need an engineer in the background. The Web3 agents are taking over all these tasks.” For how ChainAware’s agents implement these three properties, see our Web3 AI agents guide.

Why Web3 Is the Perfect Environment for AI Agents — and Web2 Is Not

One of the conversation’s most structurally important arguments concerns why AI agents will emerge primarily from Web3 rather than Web2 — and why the mainstream tech press’s framing of AI agents as a Web2 phenomenon misses the specific infrastructure advantage that Web3 provides.

The fundamental issue is data continuity. Web2 applications are built on siloed, proprietary data systems — a company’s CRM data, ERP data, customer transaction history, and operational data all live in separate systems with separate access controls, different formats, and institutional barriers to sharing. When a Web2 business process needs to flow across organizational boundaries, it invariably encounters a break: a human must intervene, data must be manually transferred, a back-office team must reconcile records, or a Business Process Outsourcing arrangement must be maintained to bridge the gap. As Martin explains: “In Web2 it is difficult to do the agents because data is missing. We have always these data breaks — silo organizations. But in Web3, we have fully digitalized data — 100% automation, which offers us the possibility that we put the agents to analyze all this data and to do these activities.” For more on how ChainAware exploits Web3’s data architecture, see our behavioral analytics guide and the Ethereum Foundation’s on-chain data documentation .

Web3 Business Processes Are 100% Digitalized

Web3 eliminates the data continuity problem entirely through the blockchain’s fundamental design. Every transaction, every state change, every protocol interaction is recorded on a shared, permissionless ledger that any agent can read without requiring access permissions, API agreements, or data sharing arrangements. A DeFi agent that needs to check a user’s lending position across five protocols, assess their collateralisation ratio, evaluate current interest rates on competing protocols, and execute a rebalancing transaction can do all of this in a single continuous operation — because all the required data exists in the same open, machine-readable format. No data silos. No process breaks. No back-office intervention. This is precisely what Martin means when he says Web3 has 100% digitalized business processes: not just that the data is digital, but that it is continuously accessible, consistently structured, and inherently cross-organisational.

Protect Users Before They Enter Any Pool

ChainAware Rug Pull Detector — 95% of PancakeSwap Pools Are at Risk

AI agents helping users find yield need to verify pool safety before any deposit. ChainAware’s Rug Pull Detector traces the contract creator’s funding chain and all liquidity provider histories to detect behavioral rug pull patterns before you invest. Free for individual pool checks on ETH, BNB, BASE, and HAQQ. Available via Prediction MCP for any agent to call programmatically.

Check Any Pool Free Rug Pull Detection Guide

The Founder Bandwidth Argument: Agents Free Humans for Innovation

Martin introduces a practical economic argument for AI agent adoption that applies directly to every Web3 founder running a project: the founder bandwidth problem. Most Web3 founders divide their time across a wide range of activities — product development, marketing, compliance, tax reporting, investment management, community management, and investor relations. The majority of these activities are not innovation. They are coordination, administration, and routine analysis that consumes enormous cognitive bandwidth and calendar space without producing the creative breakthroughs that justify founding a startup in the first place.

AI agents, applied systematically, can take over most of these supplementary functions. A marketing agent continuously generates and optimises personalised content for different user segments. A transaction monitoring agent continuously screens the platform’s user base for compliance risks. A credit scoring agent continuously evaluates borrower creditworthiness. Each of these agents performs work that would otherwise require a dedicated human specialist — but they do it 24/7, without management overhead, at a cost that scales with computing resources rather than headcount. The result, as Martin argues, is that founders regain the bandwidth to focus on what human brains are actually designed for: creating genuinely new things. As Martin explains: “Co-founders will have much more space, much more bandwidth for the innovation. Instead of dealing with marketing, compliance, bookkeeping, tax — all these supplementary activities — the agents take them over. And I think that is what human brains are created for: creating new things, creating innovations.” For how the marketing agent specifically addresses the founder bandwidth problem, see our Web3 AI marketing guide.

Every Web3 Project Is Bottlenecked on the Same Supplementary Tasks

The universality of the founder bandwidth problem across Web3 projects is itself significant. Whether a project is a DeFi lending protocol, a gaming platform, a DEX aggregator, or an analytics layer, the supplementary task load is remarkably similar: marketing to reach new users, compliance to satisfy regulatory requirements, fraud monitoring to protect the platform, and tax and accounting to manage the treasury. The specifics differ, but the categories are consistent. This means that AI agents designed to address these categories are not niche tools for specific project types — they are horizontal infrastructure that benefits every Web3 project simultaneously. For how ChainAware’s agent stack addresses these categories, see our Web3 agentic economy guide.

Trigger-Based Agents: The Building Blocks of the DeFi Agent Economy

Ayush provides a concrete starting point for understanding how DeFi AI agents operate at the basic functional level — one that helps demystify agent architecture for founders and users who are intimidated by the concept. The simplest form of a DeFi agent is a trigger-based executor: it monitors a specified condition and executes a defined action when that condition is met, without any further human involvement.

Consider a straightforward example: a user wants to buy a specific token when its price reaches $100. On a centralised exchange, a limit order handles this trivially. On DeFi platforms, the same operation is significantly more complex — spot trading at specific price points requires continuous monitoring, gas fee management, slippage handling, and often cross-protocol interaction. A trigger-based agent abstracts all of this complexity: the user specifies the condition and the action, the agent monitors continuously, and the execution happens automatically when the trigger fires. As Ayush explains: “You can just give the agent a task — if somebody can train an agent that if the market is volatile, you can tell the agent that I want to swap my USDT when the price of a certain token hits $100. So this is a very simple task but it is very difficult to do such a thing on DeFi platforms. So these kinds of initial building blocks are what we are going to utilise and then eventually we can build and make more and more complex agents.” For more on how ChainAware’s predictive models power agent decision-making, see our Prediction MCP guide.

From Simple Triggers to Complex Autonomous Strategies

The trigger-based agent is the entry point — but the architecture scales to arbitrarily complex strategies. A simple trigger monitors one condition and executes one action. A more complex agent monitors multiple conditions simultaneously (price thresholds, liquidity depth, fraud probability, collateralisation ratios), weighs them against a defined objective function (maximise yield subject to maximum risk tolerance), and executes multi-step transaction sequences across multiple protocols. The computational complexity grows rapidly, but the underlying architecture — condition monitoring, decision logic, execution — remains consistent. This is why Ayush describes trigger-based agents as “building blocks”: they are the atomic units from which arbitrarily sophisticated autonomous strategies can be assembled.

ChainAware’s Web3 AI Agents: Marketing Agents and Transaction Monitoring

Martin describes ChainAware’s two primary agent products in detail, explaining how they each address a specific high-value problem for Web3 platforms using the predictive AI and behavioral analytics infrastructure that the team has built over multiple years.

The Web3 marketing agent operates at the moment a wallet connects to a platform. At that instant, the agent retrieves the wallet’s on-chain behavioral history, calculates its behavioral profile using ChainAware’s predictive models (experience level, risk willingness, intentions — borrower, trader, staker, gamer, NFT collector), and generates content specifically matched to that profile. Borrowers see lending-focused content. Traders see leverage and position management content. NFT-oriented wallets see content connecting the platform’s features to the NFT ecosystem they already use. The entire process is fully automated — no human marketer reviews or approves individual messages. As Martin explains: “We fully automated from one side prediction, from the other side content generation. And we have now Web3 agents — a marketing agent, self-running and autonomous.” For the complete marketing agent methodology, see our AI marketing guide.

The Transaction Monitoring Agent: Compliance Simplified

The transaction monitoring agent addresses a different but equally pressing need: continuous compliance monitoring of an active user base. Under MiCA regulation and FATF Recommendation 16 , every Virtual Asset Service Provider is required to implement AI-based transaction monitoring — not just backward-looking AML fund tracking, but forward-looking behavioral analysis that identifies fraud risk before transactions occur. The transaction monitoring agent accepts a set of wallet addresses (the platform’s connected users) and monitors all of their on-chain activity continuously across every supported blockchain. When behavioral patterns emerge that match fraud signatures, the agent automatically flags the address and notifies the compliance officer via Telegram or the platform interface. As Martin explains: “Instead of having compliance departments — and soon every virtual asset service provider has to set up a compliance department — you set up transaction monitoring agents and they do this stuff. They track, they flag things if things are not okay.” For the full regulatory context, see our AML and transaction monitoring guide and our compliance guide.

The Agent-to-Agent Economy: $5-10 Billion and a Paradigm No One Fully Understands Yet

The conversation’s most forward-looking section addresses a vision that both Ayush and Martin describe with genuine intellectual humility: the agent-to-agent economy — a system where AI agents communicate directly with each other to accomplish objectives, without any human in the interaction loop.

The concept builds on current agent architectures but takes them to a logical extreme. Rather than a human defining a goal and an agent executing it, the agent-to-agent model involves one agent delegating subtasks to other agents, which may in turn delegate to further agents — all autonomously, all in real time, all optimising toward the original objective. A top-level “portfolio optimisation” agent might simultaneously query a yield-finding agent, a fraud assessment agent, a liquidity depth agent, and a gas fee optimisation agent — receiving their outputs, synthesising them, and executing a transaction sequence that no single human could have coordinated in the available timeframe. Ayush draws a parallel to the Internet of Things, which promised a similar seamless interconnection of devices: “This AI agent economy can be huge. We were expecting something similar with the Internet of Things where our appliances and electronics can talk to each other. I think this is where we are coming. And this AI agent economy is expected to be $5 to 10 billion in the next 3 to 4 years.” For context on the AI agent economy’s broader commercial potential, see Grand View Research’s AI agents market report . For how ChainAware’s Prediction MCP enables agent-to-agent querying, see our 12 blockchain capabilities guide.

Nobody Knows How Big It Will Be

Both Martin and Ayush are explicit about the limits of their forward visibility — and this honesty is itself significant. Projects that claim to have a complete roadmap for an agent-to-agent economy that does not yet exist are either deluding themselves or their investors. The honest position is that the technology convergence enabling this economy is assembled and operational, the first applications are live and demonstrating value, and the scaling trajectory is directionally clear — but the endpoint is genuinely unknown. As Martin puts it: “We do not know what is coming yet. It is like we are just starting this innovation now. Everything that we did before, we are preparing for the wave of innovation. And this innovation wave is starting.” This calibrated uncertainty is not a weakness — it is an accurate reflection of how transformative technological transitions work. The people building the early internet in 1994 could not have predicted Amazon, Google, or Netflix.

Web3 vs Web2 for Agents: Cross-Chain Open vs Android/iOS Closed

Ayush provides a concrete analogy that makes the structural difference between Web3 and Web2 for agent deployment immediately intuitive. In Web2, building an application for Android and then wanting to deploy it on iOS requires essentially building the application again from scratch — the two platforms have incompatible architectures, different development frameworks, different app store policies, and different runtime environments. Interoperability between them is limited, negotiated, and controlled by the platform owners. As Ayush observes: “In Web2, if you are building an application on Android and if you want to launch it on iOS, it is a completely new application.” Web3 does not work this way. A smart contract deployed on Ethereum can be called by any application on any chain that supports the relevant bridge or cross-chain messaging protocol. An AI agent querying ChainAware’s Prediction MCP receives behavioral data from eight blockchains through a single API call — not through eight separate integration projects with eight separate permission negotiations. The openness that is often discussed as a philosophical feature of Web3 turns out to be a specific practical enabler for AI agent deployment at scale. For how ChainAware’s multi-chain architecture enables this, see our AI agents acceleration guide.

The Convergence: Web3 + AI Models + Real-Time Data + Autonomous Operation

Martin synthesises the conversation’s key argument into a convergence framework that explains why the AI agent moment is happening now rather than three years ago or three years from now. The innovation wave requires a specific set of technologies to exist simultaneously — no single component is sufficient, and the full set only recently became available together.

Web3 provides the 100% digitalized, open, permissionless data infrastructure. AI models — both predictive (ChainAware’s behavioral classifiers) and generative (LLMs for content generation) — provide the intelligence layer. Real-time data feeds eliminate the 18-24 month latency that made early LLMs unsuitable for time-sensitive decisions. Autonomous, continuously running operation removes the human from each interaction cycle. The convergence of all four creates something qualitatively different from any of the components individually: an agent that can perceive the current state of a blockchain ecosystem, reason about it with trained intelligence, generate appropriate responses, and execute consequential actions — without requiring human initiation, monitoring, or approval at each step. As Martin explains: “We need this convergence. There has to be Web3, there has to be AI models, AI models have to be real-time — now we have this continuous approach. So we have all this convergence of different technologies which is possible in Web3 only, not in Web2. And this economic impact is huge.” For how ChainAware’s architecture reflects this convergence, see our real AI use cases for Web3 guide and refer to McKinsey’s State of AI report for broader convergence trends.

Data Privacy and AI Agents: The Matrix Analogy and the User’s Choice

An audience question during the X Space raises data privacy — a concern that applies to any system that processes behavioral data about individuals. For AI agents that analyse on-chain transaction histories, the privacy question has a specific and interesting structure: blockchain data is inherently public, yet the behavioral profiles derived from it can be deeply personal.

Both Martin and Ayush address this from different angles, arriving at a shared conclusion: data privacy in Web3 AI agents is primarily a matter of user choice rather than a system design limitation. Martin’s perspective is grounded in a simple trade-off: users who share their wallet history with ChainAware’s agents receive the most relevant, personalised experiences and the most useful ecosystem interactions. Users who prefer privacy can use fresh addresses with no transaction history — they will receive default generic experiences rather than personalised ones, but their privacy is fully preserved. As Martin explains: “Some people don’t want to expose the data. People who want to expose the data will use their wallets. Others will use empty wallets. Now if people are using their data, this data is the best business card — you know, can you trust them, what are their intentions, what is their experience?” For how the Wallet Auditor implements this trade-off in practice, see our wallet auditor guide.

The Matrix Analogy: Seeing the Person Behind the Blockchain Data

Martin uses the Matrix film as a reference point to describe two fundamentally different ways of perceiving blockchain data — and by extension, two fundamentally different capabilities for building agents that interact meaningfully with blockchain users. The analogy is precise and illuminating.

In the Matrix, some characters see the screen of cascading green characters — the raw data stream of the simulation. Others — like Neo after his awakening, or the veteran operator Tank — see through the characters to the objects and people they represent. The two groups are looking at the same data but perceiving entirely different realities. Blockchain data presents the same dual perception possibility. At the surface level, it is a stream of cryptographic hashes, addresses, and transaction amounts — opaque to most users and requiring significant technical knowledge to interpret at all. At the deeper level, it is a rich record of human financial behavior: risk preferences, experience levels, protocol loyalties, intention patterns, and social connections — all permanently recorded and available to anyone with the analytical tools to extract them. As Martin explains: “Like a character, Spitts and bites at the screen — other people like Neo see the persons behind the green characters on the screen. Like some people are maybe now focusing on the data privacy and so but it’s — everyone can decide himself. If somebody is very data privacy centric, use always a new address. But it means you will get less impact, less output from the Web3 ecosystem.” For how ChainAware’s behavioral analytics platform makes this deeper perception operationally accessible, see our behavioral analytics guide and our Web3 business intelligence guide.

See the Person Behind the Blockchain Data

ChainAware Wallet Auditor — Full Behavioral Profile in 1 Second

Stop seeing only the address. Start seeing the person: experience level (1-5), risk willingness, predicted intentions (trader, borrower, staker, gamer), fraud probability, and Wallet Rank. Free, no signup, instant results. The tool that powers ChainAware’s agent stack — available to any user or developer.

Audit Any Wallet Free Wallet Auditor Guide

Comparison Tables

Web2 vs Web3 as AI Agent Deployment Environments

AI Agent Types in Web3: What They Do, Who Benefits

Agent Type	What It Does	Who Benefits	Status
Marketing Agent (ChainAware)	Calculates wallet behavioral profile at connection, generates 1:1 resonating content automatically	DApp founders — reduces CAC, increases conversion	Live — GTM 2-line integration
Transaction Monitoring Agent (ChainAware)	Continuously monitors platform user addresses, flags fraud patterns, alerts compliance via Telegram	DApp compliance teams — expert-level 24/7 monitoring	Live — subscription
Yield Optimisation Agent	Finds best yield across protocols, chains, tokens — executes rebalancing automatically	Token holders — removes complexity of DeFi navigation	Emerging — UniLend LLAMA, others
Trigger-Based Trading Agent	Executes swap/position actions when specified price/condition triggers are met	Traders — automates condition-based DeFi execution	Emerging — initial building blocks
Research & Alpha Agent	Finds new tokens, evaluates fundamentals, identifies market opportunities	Retail investors — replaces manual research across dozens of sources	Emerging — early tools available
Fraud Detection Agent (ChainAware)	Evaluates wallet fraud probability before any interaction — 98% accuracy, real-time	Users + protocols — prevents losses before they occur	Live — free for individuals, API/MCP for businesses
Credit Scoring Agent (ChainAware)	Calculates on-chain creditworthiness for DeFi lending decisions	Lending protocols — enables under-collateralised lending	Live on ETH — broader demand in 12-24 months
Compliance Agent	Automated MiCA/FATF compliance monitoring, reporting, and flagging	VASPs — removes compliance department headcount requirement	Live (ChainAware TM Agent) + broader market developing

Frequently Asked Questions

What is UniLend Finance and what is the LLAMA platform?

UniLend Finance is a DeFi protocol live on blockchain since 2021, offering permissionless lending and borrowing — any token can be listed for lending and borrowing without governance approval, analogous to how any token can be listed on Uniswap for trading. UniLend V1 has approximately $4.2 million in TVL and V2 extends the permissionless model. LLAMA is UniLend’s upcoming platform for launching AI agents on blockchain — designed to let anyone build and deploy task-oriented agents without machine learning expertise or agent development experience. The platform specifically focuses on agents that complete real tasks rather than just producing conversational outputs, with hackathons and community programs planned around it.

Why do most token holders never use DeFi, and how do AI agents fix this?

Approximately 95% of crypto token holders never use DeFi lending, borrowing, or yield optimisation products — despite owning assets that could be generating passive income. The barriers are practical: navigating multiple chains and protocols, evaluating security risks, managing gas fees, understanding liquidation mechanics, and monitoring positions continuously requires significant expertise and time investment. AI agents remove every one of these barriers by handling the full process autonomously. A user specifies a goal (earn yield on USDC, minimise risk), and the agent finds the best protocol, evaluates its safety using fraud and rug pull detection, executes the deposit, and monitors the position — without the user needing any protocol knowledge or ongoing attention.

What makes Web3 a better environment for AI agents than Web2?

Web3’s 100% digitalized, openly accessible data architecture eliminates the data continuity problem that prevents AI agents from operating autonomously in Web2 environments. Web2 data lives in proprietary silos — a company’s CRM, ERP, and transaction systems are separate, access-controlled, and require API agreements and manual reconciliation at every organisational boundary. Every business process that crosses a boundary requires human intervention. Web3 eliminates these boundaries entirely: all on-chain data is public, permissionless, and consistently formatted. An agent can read a user’s complete DeFi history across eight chains and fifty protocols in a single query, execute a cross-protocol rebalancing transaction, and comply with regulatory reporting requirements — all in one autonomous operation, with no human in the loop.

What is the agent-to-agent economy and when will it arrive?

The agent-to-agent economy is a system where AI agents communicate directly with each other to accomplish objectives, without human mediation at each interaction. A portfolio optimisation agent, for example, might autonomously query a yield-finding agent, a fraud assessment agent, a liquidity depth agent, and a gas fee agent — synthesise their outputs — and execute a multi-step DeFi strategy, all without any human involvement beyond the initial goal specification. The market for AI agent infrastructure is expected to reach $5-10 billion within 3-4 years. Both Martin and Ayush acknowledge that nobody fully understands the endpoint yet — the honest position is that the enabling technology convergence is now in place and the building blocks are being assembled, but the full scope of what emerges will surprise even the builders.

How does ChainAware handle data privacy in its AI agent products?

ChainAware’s agent products operate on publicly available on-chain transaction data — they do not require users to submit any personal information, create accounts, or consent to data collection beyond what is already public on the blockchain. Users who want maximum personalisation from ChainAware’s marketing agents and behavioral profiles share their real wallet address, which gives the agents access to their full transaction history. Users who prioritise privacy can interact using fresh addresses with no transaction history — they receive generic default experiences rather than personalised ones, but no behavioral data is exposed. The privacy trade-off is therefore entirely user-controlled: more data shared results in more useful agent interactions; less data shared results in less personalisation but full privacy preservation.

This article is based on the X Space between ChainAware.ai co-founder Martin and Ayush from UniLend Finance. Listen to the full recording on X . For integration support or product questions, visit chainaware.ai.

The post Revolutionizing Web3 with AI Agents first appeared on ChainAware.ai.

X Space with ChainGPT and Datai — ChainAware co-founders Martin and Tarmo join Ellie from Datai and ChainGPT Labs host Chris for a wide-ranging conversation on AI agents in Web3: what they actually are, what they can already do, and why they mark the beginning of the biggest innovation wave the industry has ever seen. Listen to the full recording on X

Three projects at the frontier of Web3 AI infrastructure sit down to talk honestly about what is actually being built. ChainAware brings two production-ready AI agents — a fraud detection agent and a Web3 marketing agent — built on proprietary predictive models trained over two years. Datai brings three years of blockchain data aggregation and a smart contract categorization AI that translates raw on-chain transactions into the behavioral language that intelligent agents need to function. ChainGPT Labs, which incubates both, provides the ecosystem context that connects these tools to the broader question every Web3 builder faces: how do you get real users, build sustainable revenue, and focus on the innovation that actually matters? Together, they map out why AI agents are not a hype narrative — they are the infrastructure layer that finally makes Web3 businesses viable.

Three Projects, One Mission: What ChainAware, Datai, and ChainGPT Are Building

ChainGPT Labs brought together two of its incubated projects — ChainAware and Datai — for this X Space precisely because their work is complementary. Both teams identified the same fundamental gap in Web3 infrastructure from different directions, and both arrived at AI agents as the solution. Understanding what each brings to the table clarifies why the combination matters.

ChainAware is a prediction engine. Starting from SmartCredit’s DeFi lending platform, Martin and Tarmo built iteratively: credit scoring required fraud detection, fraud detection extended to rug pull prediction, behavioral modeling followed, and marketing personalization emerged from behavioral data. Today the platform produces real-time behavioral profiles for any wallet address — predicting fraud probability, rug pull risk, experience level, risk tolerance, and future behavioral intentions (borrower, lender, trader, gamer, NFT collector). Two production AI agents sit on top of that infrastructure: the fraud detection agent and the Web3 marketing agent. As Martin explains: “We are a big calculation engine. Not just a calculation engine — we are a prediction engine. We predict what wallets are doing in the future.” For the complete ChainAware architecture, see our product guide.

Datai: Making Blockchain Data Readable for AI

Datai approaches the same problem from the data infrastructure layer. Ellie explains the core challenge: when you look at any blockchain transaction explorer, you see addresses interacting with other addresses. However, you do not see what the user was doing. That address could be connecting to a DeFi lending protocol, minting an NFT, bridging assets between chains, signing a contract, purchasing a gaming asset, or investing in a real-world asset. The transaction looks identical at the address level regardless of which of these activities is occurring. Datai spent three years manually aggregating blockchain data and building categorization for the smart contracts that users interact with — then invested 1.5 years building an AI model that can automatically categorize smart contracts at scale. The result is data that, as Ellie puts it, reads “like English” — structured behavioral context that AI agents can actually understand and act on. For how clean behavioral data enables better AI agent decisions, see our behavioral analytics guide.

What AI Agents Actually Are: Beyond the Hype

The X Space opens with an accessible definition that cuts through the significant volume of AI agent hype circulating in the Web3 space. AI agents are autonomous systems that run continuously, learn from feedback, and execute defined functions without requiring human initiation at each step. They differ from chatbots and simple automations in three specific ways: they operate on real-time data rather than static training sets, they learn continuously from outcomes rather than remaining fixed, and they execute consequential actions (transactions, content generation, risk flags) rather than just producing text responses.

Ellie offers the most accessible definition in the conversation: “Just a friend. Like it’s a robot friend who’s living inside your PC. This robot friend will listen to what you say, what you do, and then it will start telling you things — find my best pictures, find my best song. It can understand a lot of information really quickly. It’s like having a super helper that is always ready.” This analogy captures the operational reality well: an agent that has been configured for a specific task runs in the background, continuously analyzing the information relevant to that task and taking defined actions when conditions are met. No human needs to ask it to start or tell it when to act. For more on how AI agents differ from prompt engineering, see our Web3 AI agents guide.

Why Web3 Is the Ideal Environment for AI Agents

Both Ellie and Martin make a specific structural point about why Web3 enables AI agents more powerfully than Web2. In Web2, building agents is technically simpler because the data is in natural language — tweets, messages, Netflix viewing history, search queries. However, that data is locked behind proprietary APIs, fragmented across closed platforms, and requires individual permission agreements with each company. Web3’s data is structurally different: every transaction is public, every interaction is permanently recorded on open ledgers, and no permission is required to read any of it. The challenge in Web3 is not access — it is interpretation. Raw blockchain data is not readable without smart contract categorization. Once that categorization layer exists (which is what Datai provides), the behavioral signal quality is dramatically superior to anything Web2 has — because every transaction represents a real financial decision with real cost attached. For how this connects to ChainAware’s behavioral prediction models, see our generative vs predictive AI guide.

The Founder Bandwidth Problem: Why 90% of Time Goes to the Wrong Things

One of the most practically resonant arguments in the entire conversation comes from Tarmo’s opening on what AI agents mean for Web3 founders. The observation is simple and verifiable by anyone who has run a startup: the actual innovation a founder set out to build receives a small fraction of their working time. The rest goes to the operational overhead that every business requires — marketing, sales, compliance monitoring, tax reporting, transaction auditing, customer support, legal coordination. None of these activities are the core innovation. All of them are essential. Together, they consume the majority of a founder’s calendar.

Tarmo frames this precisely: “Just imagine when you are doing now a startup. You can spend maybe a real innovation for a small piece of time. The rest of time goes into tax reporting, into marketing, into sales, into transaction monitoring. What AI agents do — they take over all these tasks which you have to do supplementary to the real innovation, so that you can focus on the innovation.” Martin reinforces this with a specific observation about Web3 marketing: most founders end up devoting enormous energy to mass marketing campaigns that produce poor conversion because the personalization infrastructure does not exist yet. Building that infrastructure, running it, and optimizing it manually consumes resources that should be going toward product iteration. For more on how marketing agents specifically address the founder bandwidth problem, see our AI marketing guide and our Web3 agentic economy guide.

The Innovation Multiplier Effect

The second-order argument is even more significant than the immediate bandwidth gain. If AI agents remove the supplementary task burden from every Web3 founder simultaneously, the aggregate increase in innovation output across the entire ecosystem is enormous. Currently, thousands of talented teams spend the majority of their time on activities that provide no competitive differentiation — mass marketing to undifferentiated audiences, manually configuring compliance monitoring, preparing tax reports. All of this effort produces zero innovation. Redirecting even half of that effort toward core product development would compound into a wave of new capability that Martin describes as the biggest the industry has seen: “This will be a massive wave of innovation that is coming. All these supplementary activities — what the founders have to do at the moment — it blocks their time. Take it over with agents. That means focus on innovation, create real innovation.” For how this connects to the broader Web3 growth trajectory, see our AI agents acceleration guide.

Deploy Your First Agent in Minutes

ChainAware Free Analytics — Know Your Real Users in 24 Hours

Before you can personalise content, you need to understand who is actually visiting your platform. ChainAware Analytics gives you the real behavioral distribution of connecting wallets — experience levels, risk profiles, intentions — in 24-48 hours. Two lines of Google Tag Manager code. Free forever. The starting point for every agent deployment.

Get Free Analytics Analytics Guide

The Web3 Marketing Agent: From Mass Messaging to 1:1 Personalization

Marketing was personalized before it became mass. Before broadcast advertising, before mass media, before the internet — merchants knew their customers individually, knew their needs, and tailored their communication accordingly. Mass marketing was an economic compromise: reaching millions of people with identical messages was cheaper per impression than reaching each person with a relevant one, even though conversion rates were dramatically lower. The internet initially intensified mass marketing rather than solving it, because the data layer needed for personalization at scale did not exist yet.

Google changed that equation in Web2 by using search and browsing history to infer behavioral intent and serve matched advertising. Web3 today sits at the same pre-AdTech position that Web2 occupied before Google’s innovation. Every major marketing channel — KOL promotions, crypto media banners, Telegram ads, CMC listings — delivers identical messages to heterogeneous audiences. A DeFi native with five years of sophisticated protocol usage receives the same onboarding content as someone who created their first wallet last week. The conversion rate from this misalignment is predictably terrible. As Martin explains: “What is website’s role? Website’s role is to convert users. Website’s role is to resonate with users. So you have to create personalized websites.” For the full Web3 personalization framework, see our Web3 personalization guide and our intention-based marketing guide.

How the Marketing Agent Works in Practice

ChainAware’s marketing agent operates at the moment a wallet connects to a platform. The sequence is: wallet connects → ChainAware’s behavioral models calculate the wallet’s profile in real time → the agent generates content matched to that profile → the visitor sees messaging that resonates with their specific behavioral type. A high-probability borrower arrives at a lending platform and sees content about borrowing terms and collateral optimization. A leverage trader at the same platform sees content about position management and leverage tools. A first-time DeFi user sees content that addresses their onboarding needs. None of these visitors know that the content was generated for them specifically — they simply experience a platform that feels relevant. As Martin explains: “You calculate the user’s behavior, experience, risk willingness. You calculate who are the future borrowers with probabilities, who are the future lenders, who are the future leverage takers, who are the gamers, who are the NFT collectors. Based on these behavioral parameters, it’s automated targeting.” For the complete marketing agent implementation, see our Web3 personas guide.

From Manual Configuration to Auto-Generation

ChainAware’s banner system — which delivers personalized messages to platform visitors based on behavioral profiles — is already in production with clients. Currently, the system includes a significant manual configuration step: a team member specifies which messages should appear for which behavioral profiles, designs the content variants, and sets the targeting parameters. This manual configuration creates a startup cost for each new client deployment. The next evolution underway is auto-generation: the agent itself generates the content variants based on the behavioral profiles it identifies, requiring only human review rather than human creation. As Martin notes: “We have a lot of manual configuration there. What we are doing now is we are moving from manual configuration to auto generation.” Once auto-generation is complete, deploying the full personalization system requires minimal setup time — and the agent runs continuously from that point without ongoing human involvement.

The Orchestrator Shift: How Marketers Evolve in an AI Agent World

The host Chris, who works in marketing and community management for ChainGPT Labs, asks the question that many marketing professionals privately wonder: do AI agents replace the marketer? The answer from both Ellie and Tarmo is thoughtful and specific — and it reframes the question in a way that is both reassuring and clarifying.

Ellie’s observation is precise: AI agents in Web3 marketing will make the marketer’s work “a bit similar to Web2.” The comparison is apt. In Web2, sophisticated marketers do not write every word of copy, design every visual, or manually A/B test every subject line — they use tools, platforms, and workflows that handle execution while the marketer focuses on strategy, brief writing, and judgment about what is and is not resonating. Web3 marketing currently operates below that level because the data layer and personalization infrastructure do not yet exist. AI agents bring Web3 marketing up to Web2 sophistication, and then push further toward genuine 1:1 personalization that Web2 never fully achieved. For the marketing professional, the transition is from manual execution to strategic orchestration. As Tarmo describes the shift: “You become like an orchestrator. You have highly specialized agents — one agent is preparing nice illustrations which resonate with specific personas, one agent is preparing your texting, one agent is calculating a psychological profile. All you do is orchestrate them.” For more on how this orchestration model works in practice, see our Web3 growth guide.

High-Value Creation vs Low-Value Execution

The practical consequence of the orchestrator shift is a redistribution of human cognitive effort from low-value execution tasks toward high-value creative and strategic work. Currently, a significant portion of any marketing team’s time goes to tasks that require skill to do but that produce no strategic differentiation: writing variations of the same message for different channels, manually segmenting audience lists, resizing images for different ad formats, reporting on campaign performance. These tasks require time and training but not genuine creative judgment. AI agents can execute all of them. What they cannot replace is the judgment about which message strategy actually resonates with a specific community, which product narrative builds genuine trust, and which creative approach communicates a technical value proposition clearly. As Tarmo explains: “We are taken out of these daily operating activities where we spend 90% of our time. Instead we focus on these high, very high value creation activities. We use our creativity, our intellectual power to create something new.” For more on how ChainAware’s agent stack supports this reallocation, see our DeFi onboarding guide.

Datai: The Data Layer That Makes Intelligent Agents Possible

For an AI agent to make intelligent decisions, it needs to understand the context of the data it is acting on. In Web2, context is relatively accessible: user behavior is expressed in natural language — search queries, messages, reviews, social posts. AI systems trained on language can interpret this behavior without additional translation layers. In Web3, the equivalent behavioral data is expressed in a format that is opaque by default: hexadecimal addresses interacting with hexadecimal contracts, with transaction values in token units. None of this raw data tells you what the user was doing in any meaningful behavioral sense.

Datai’s core product solves this interpretation problem. By categorizing the smart contracts that users interact with, Datai transforms raw transaction histories into behavioral narratives. A series of transactions that looks like “0x4f…a2 interacted with 0x7d…c8” becomes “this wallet borrowed USDC on Aave, provided liquidity on Uniswap, bridged to Arbitrum, and purchased a gaming asset on Immutable X.” That translated narrative is what Ellie means by data that reads “like English” — structured, categorized behavioral context that AI agents can process, segment, and act on without requiring custom interpretation for each new protocol or chain. As Ellie explains: “When a user is interacting with a smart contract, there can be a thousand ways of what they’re doing — connecting to a DeFi protocol, interacting with NFT, bridging, signing a contract, maybe buying a gaming asset, investing in real world assets. If you look at the scanner, you see only addresses. But what are those addresses? What is the user doing? This is exactly what we’re trying to solve.” For how ChainAware’s models use behavioral data, see our blockchain analysis guide.

Smart Contract Categorization: Translating Addresses into Behavior

The practical value of smart contract categorization becomes clear when you consider the analytics problem any DApp operator faces. A platform operator knows everything about what users do inside their own protocol — how much liquidity they add, how long they stay, what assets they prefer. However, they know nothing about what those same users do everywhere else on the blockchain. A lending platform does not know whether its users also trade on derivatives protocols, whether they are active NFT collectors, whether they bridge frequently to other chains, or whether they have significant capital sitting idle in other protocols that they might potentially move. All of that behavioral context exists in public blockchain data — it is simply not interpretable without the categorization layer that tells you what each smart contract interaction represents.

Datai’s categorization layer makes this cross-platform behavioral picture available. As Ellie explains: “We can tell you that 10% of your customers are using lending-borrowing platforms on the same chain or on different chains. What assets are they lending and borrowing that you don’t have internally? So you can adjust your product strategy based on the behavior of what your customers are doing outside of the platform.” This external behavioral view is the Web3 equivalent of Google Analytics combined with competitor research — understanding not just what users do on your platform but who they are in the broader behavioral ecosystem. For how ChainAware’s wallet auditor provides a similar behavioral picture for individual wallets, see our wallet auditor guide and our user segmentation guide.

The Fraud Detection Agent: Protecting the Ecosystem, Not Just One Platform

Martin frames ChainAware’s fraud detection agent not as a product that protects individual users, but as ecosystem infrastructure that affects whether Web3 grows at all. The argument connects directly to the new user retention problem: every time a new participant enters Web3 and encounters a rug pull or scam, there is a meaningful probability they leave permanently. They do not distinguish between one bad project and the broader ecosystem — they associate the negative experience with the entire space and return to centralised exchanges or exit crypto altogether. Experienced participants — the OGs Martin refers to — have developed instincts for avoiding the worst situations. But new users have not.

The scale of the fraud problem in DeFi is significant. ChainAware’s data on PancakeSwap pools is striking: 95 to 98% of new pools end in rug pulls. That number means the base rate expectation for a new user exploring DeFi liquidity provision is almost certain loss. No amount of excellent UX or product innovation can overcome a user experience where the majority of initial interactions result in total loss of funds. Reducing that fraud rate — not just for individual users but across the ecosystem — is therefore a prerequisite for Web3 mainstream adoption. As Martin states: “It’s not just for one person, it’s not just for one DApp — it’s for the full ecosystem. If you clean up the ecosystem, we increase the trust, we get much more users, we get much more usage.” For the complete fraud detection methodology, see our fraud detection guide and our fraud detector guide.

Free Tools as Ecosystem Infrastructure

ChainAware’s decision to offer fraud detection and rug pull detection tools free to individual users reflects this ecosystem logic directly. If the goal were purely commercial, these tools would be paywalled to maximize revenue per user. The actual goal, however, is ecosystem trust improvement — which requires maximum adoption. Every user who checks an address before interacting with it, and every user who avoids a rug pull because they checked the pool contract, represents one fewer negative experience that might have driven a new participant out of Web3 permanently. At scale, widespread adoption of free fraud detection tools changes the ecosystem-level new user retention rate. For the free tools, see our fraud detector guide and our rug pull detection guide. For context on crypto fraud scale, see Chainalysis’s annual crypto crime data .

Protect Your Users Before Any Interaction

ChainAware Fraud Detector + Rug Pull Detector — 98% Accuracy, Real-Time, Free

95-98% of new DeFi pools end in rug pulls. 98% of fraud can be predicted before it happens. Enter any wallet address or contract and get a real-time behavioral risk score — backtested on CryptoScamDB. Half a second for standard addresses. Free for every user on ETH, BNB, BASE, and HAQQ.

Check Fraud Risk Free Rug Pull Detector

Transaction Monitoring Agent: The Regulatory Requirement That Protects Everyone

Beyond the individual user tools, ChainAware’s transaction monitoring agent serves a specific regulatory function for platform operators. Under MiCA regulation and FATF recommendations, Virtual Asset Service Providers — which includes most DeFi protocols — must implement both AML analysis and AI-based transaction monitoring. These are not the same thing, and Martin is precise about the distinction throughout the conversation.

AML analysis is a rules-based system that tracks the flow of known-illicit funds through the blockchain. It is inherently backward-looking and static: it can only flag addresses connected to previously identified fraud. Transaction monitoring, by contrast, uses AI to analyze behavioral patterns in real time and predict which currently legitimate-appearing addresses are likely to commit fraud in the future. The operational difference matters because sophisticated fraud operations design their activity specifically to pass AML checks while their behavioral history already contains the patterns that predictive AI identifies. As Martin explains: “Scammers and hackers — it’s a dynamical system. You cannot go with rules against a dynamical system. You need AI to interact with this dynamical system. That’s why you need transaction monitoring.” For the full regulatory context, see our AML and transaction monitoring guide and the FATF virtual assets recommendations .

The Transaction Monitoring Agent in Operation

The operational model for the transaction monitoring agent is straightforward to implement. A platform operator uploads a list of wallet addresses — the connected users of their protocol — ranging from a few hundred to several thousand. The agent monitors all of these addresses continuously across all supported blockchains. When behavioral patterns emerge that match the fraud signature library (patterns that have historically preceded fraudulent activity, even in addresses that have not yet committed visible fraud), the agent flags the address and notifies the relevant compliance contact via Telegram or the platform interface. The compliance officer then makes the decision about what action to take — shadow restriction, investigation, or automated exclusion. The human remains in the decision loop, but the detection and notification happens automatically, continuously, without any ongoing human monitoring effort. For the complete transaction monitoring implementation, see our transaction monitoring guide.

Datai’s Trading Use Case: From Pre-Packaged Strategies to Personalized AI Agents

Ellie’s description of Datai’s trading AI agent use case traces a clear evolutionary arc in how DeFi users interact with complex financial strategies. DeFi began as a series of raw protocol interactions — users manually navigating Aave, Uniswap, Compound, and other protocols to construct their own yield strategies. In 2020, platforms began packaging these interactions into pre-built strategies: users could select from a menu of two to ten defined approaches, each representing a different combination of protocols, assets, and risk parameters. This was an improvement, but it created a different problem: the strategies were designed for generic user profiles, not for individual behavioral histories.

A user who primarily trades stable pairs and never touches leveraged positions faces the same menu of strategies as a user who actively manages high-risk leveraged portfolios across multiple chains. Neither user gets a strategy actually calibrated to their risk tolerance, behavioral history, or current asset holdings. The AI agent approach changes this entirely. As Ellie describes: “Wallet providers are developing agents that will go and analyze all your trading history — did you trade meme coins, stablecoins, add liquidity, borrow, leverage yourself? Based off this deep understanding, they create strategies that are fit to the user’s behavior.” The agent additionally considers what other users with similar behavioral profiles have done — a peer comparison layer that makes the recommendation more robust than individual history alone. For more on how behavioral profiling enables this personalization, see our behavioral analytics guide.

The Pool Comparison Product: A Practical Agent Application

Ellie shares a concrete product example that illustrates how data infrastructure enables AI agent functionality. Datai built an internal tool that tracks a single liquidity pool (for example, ETH/USDT) across all major protocols — Uniswap, Sushiswap, PancakeSwap, and others — comparing APY performance, liquidity depth, and security parameters simultaneously. A crypto fund initially used this to track their own portfolio performance. Then an external company building a trading AI agent contacted Datai to integrate this data: the agent needed to know which version of a given pool across which protocol and chain offered the best combination of yield and security at any given moment, then use bridging to route the user’s capital to the optimal destination automatically. As Ellie explains: “You want to invest in the same pool. You have maybe 100 possibilities. AI agents are built to help you better guide your choices. You just say: I want to add ETH/USDT to a pool. I don’t care if I’m on Ethereum or Base. It’s funneled to the right chain and the protocol with acceptable liquidity and highest APY.” For a parallel example using ChainAware’s Prediction MCP for agent decision-making, see our Prediction MCP guide.

The Web2 Parallel: Two Technologies That Drove the Crossing of the Chasm

Both ChainAware and Datai converge on the same historical framework for understanding Web3’s current position. The Web2 internet went through an identical phase before mainstream adoption: a technically sophisticated early-adopter community, significant innovation in business process efficiency, but brutal user acquisition costs driven by mass marketing and a persistent trust problem driven by widespread fraud. Web2 crossed from niche to mainstream through two specific technological interventions — and both Martin and Ellie name them explicitly.

The first was fraud detection. Credit card fraud was so pervasive in Web2’s early commercial phase that consumer reluctance to transact online constrained the entire e-commerce sector. Web2 companies collectively spent enormous development resources fighting fraud before they could focus on growth. The solution was transaction monitoring systems — mandated by financial regulators for payment processors, implemented in AI-based real-time pattern detection. Once fraud rates dropped, consumer trust increased and new users stopped burning their fingers and leaving. Ellie frames this directly: “Web2 became real. Web2, before what we know now, developed two very important technologies. One of them was fraud detection. It was fighting of credit card fraud.” For the complete historical parallel, see our ChainAware vs Google Web2 guide.

AdTech: The Second Technology That Made Web2 Viable

The second technology was AdTech. Before Google’s innovation, Web2 marketing was mass marketing — banner ads, email blasts, and press releases that reached everyone identically regardless of intent. Customer acquisition costs were prohibitively high because undifferentiated messages produced low conversion rates. Google used search history and browsing behavior as a proxy for intent, combined micro-segmentation with targeted delivery, and reduced customer acquisition costs from thousands of dollars to tens of dollars. Twitter, Facebook, and every major Web2 platform followed with their own behavioral targeting systems. The business models that power the modern internet — $600+ billion annually in digital advertising — exist because AdTech made user acquisition economically viable. As Ellie summarises: “The second crucial technology that Web2 had before it became mainstream was AdTech. Web2 used AdTech to match in an invisible way buyers and sellers. These were two key technologies which were the basis of our current Web2 world.” For AdTech scale data, see Statista’s Google advertising revenue data . For how ChainAware replaces Google’s role in Web3, see our Web3 AdTech unit costs guide.

Web3 Is at the Same Inflection Point

Web3 today mirrors Web2 at the pre-chasm moment almost exactly. There is a sophisticated early-adopter community, significant innovation in business process automation (unit costs of financial operations have fallen dramatically), persistent fraud that drives new users away, and catastrophic user acquisition costs driven by mass marketing that does not convert. The two solutions that worked in Web2 — AI-based fraud detection and behavioral targeting AdTech — are now available for Web3 in a form that is structurally superior to what Web2 had, because blockchain transaction data carries higher behavioral signal quality than search history. As Martin concludes: “It happened because the fraud was taken down in the ecosystem. And from the other side, the crossing was introduced by Google. Google was the innovator. Now we are in Web3, exactly in the same situation as Web2 once was. How do we cross the chasm? Reduce fraud. Bring in personalized AdTech.” For more on how this two-part solution maps to ChainAware’s product roadmap, see our Web3 growth guide and Geoffrey Moore’s Crossing the Chasm framework .

The Coming Innovation Wave: What Happens When Founders Get Their Time Back

The conversation closes with both Martin and Tarmo making a forward-looking argument that goes beyond the near-term benefits of individual AI agent deployments. The second-order effect of AI agents removing supplementary task burdens from every Web3 founder simultaneously is not incremental improvement — it is a step-change in the industry’s aggregate innovation capacity.

Currently, the Web3 ecosystem contains thousands of technically capable teams building genuinely novel infrastructure. Most of them spend the majority of their working time on activities that require skill but produce no differentiation — the same mass marketing campaigns, the same compliance monitoring procedures, the same administrative overhead. When AI agents absorb those tasks, the collective human creative capacity that was previously consumed by execution gets redirected toward product ideation, architectural decisions, and genuine innovation. Tarmo’s framing is direct: “With AI agents in marketing, AI agents in trust systems and fraud detection, we can bring the entire Web3 ecosystem to a new level.” This is not a marginal improvement to existing trajectories — it is a qualitative shift in what Web3 can produce. For context on the AI agent economy’s growth trajectory, see the Grand View Research AI agents market report and our real AI use cases guide.

Deploy the Full Agent Stack

ChainAware Prediction MCP — 18M+ Personas, 8 Blockchains, 32 Open-Source Agents

Every ChainAware capability — fraud detection (98%), rug pull prediction, behavioral profiling, marketing personalization, transaction monitoring — accessible via a single Prediction MCP. Any AI agent queries it in natural language and gets real-time behavioral predictions. 32 MIT-licensed agents on GitHub. SSE-based integration in minutes.

Get MCP Access View 32 Agents on GitHub

Comparison Tables

ChainAware vs Datai: Complementary AI Agent Infrastructure Layers

Pre-Packaged DeFi Strategies vs AI Agent Personalized Strategies

Frequently Asked Questions

What is ChainGPT Labs and why did it incubate both ChainAware and Datai?

ChainGPT Labs is the incubation and investment arm of ChainGPT, a blockchain-focused AI platform and IDO launchpad. The incubation thesis focuses on projects building real AI infrastructure for Web3 — specifically those with proprietary technology, genuine use cases, and measurable product traction rather than narrative-driven projects. Both ChainAware and Datai fit this thesis: ChainAware with its proprietary predictive AI models (fraud detection, rug pull prediction, behavioral profiling) and Datai with its three-year smart contract categorization dataset and AI model. The X Space brought both together specifically because their capabilities are complementary — ChainAware predicts future wallet behavior while Datai provides the historical behavioral context that makes predictions richer and more accurate.

How does ChainAware’s marketing agent protect user privacy?

ChainAware’s marketing agent operates exclusively on publicly available on-chain transaction data. No personal identity information is required at any point. When a wallet connects to a platform, the agent calculates a behavioral profile from that wallet’s public transaction history — experience level, risk tolerance, intentions — and generates matched content accordingly. The user remains fully anonymous throughout: the agent knows behavioral patterns but not personal identity. This means the personalized experience is delivered without any KYC process, without cookie tracking, and without any data that could identify the individual behind the address. As Martin notes in the conversation: “Anonymity is still there, but we know the behavior of a person behind this address.”

What problem does Datai solve that wallet analytics tools do not?

Standard wallet analytics tools show you what transactions a wallet executed — the addresses it interacted with, the values transferred, the timing. They do not tell you what the wallet was doing in any behavioral sense. A wallet that interacted with 0x4f…a2 could have been borrowing USDC, providing liquidity, bridging ETH, or purchasing an NFT — the address looks identical in all cases. Datai’s smart contract categorization layer solves this interpretation problem by mapping every smart contract address to its functional category and behavioral context. The result is that wallet transaction histories become readable behavioral narratives: “this user borrowed on Aave, traded on Uniswap, bridged to Arbitrum, and purchased a gaming asset” — context that AI agents can act on meaningfully.

Will AI agents replace Web3 marketing professionals?

The consensus from both ChainAware and Datai is no — but the role changes significantly. AI agents take over execution tasks: generating content variants, segmenting audiences by behavioral profile, serving personalized messages, monitoring campaign performance, and optimizing targeting parameters. What they do not replace is strategic judgment: deciding which product narrative builds genuine community trust, identifying which behavioral segments represent the highest-value users, designing the creative brief that agents execute from, and evaluating whether the overall strategy is achieving its goals. The marketer becomes an orchestrator of specialized agents rather than a manual executor — which is, as Ellie notes, similar to how sophisticated Web2 marketing professionals already work with marketing technology platforms today.

What is the crossing the chasm requirement for Web3 mainstream adoption?

Both ChainAware and Datai identify the same two requirements, directly parallel to what drove Web2’s crossing of the chasm. First, fraud rates must decrease significantly through widespread deployment of AI-based fraud detection — making the ecosystem safe enough for new users to stay and build positive experiences rather than burning their fingers and leaving permanently. Second, user acquisition costs must drop from the current ~$1,000 per transacting DeFi user to something closer to Web2’s $15-30 benchmark — achievable through behavioral targeting AdTech that replaces mass marketing with intent-matched personalization. Both ChainAware’s production agents and Datai’s data infrastructure directly address both requirements. When both are solved simultaneously, the conditions for mainstream adoption are in place — exactly as they were when Web2 deployed transaction monitoring and AdTech in the early 2000s.

This article is based on the X Space hosted by ChainGPT Labs featuring ChainAware co-founders Martin and Tarmo alongside Ellie from Datai. Listen to the full recording on X . For integration support or product questions, visit chainaware.ai.

The post AI Agents in Web3: From Hype to Production Infrastructure — X Space with ChainGPT and Datai first appeared on ChainAware.ai.