In March 2026, Illia Polosukhin, co-founder of the NEAR protocol, made a seemingly simple yet profound statement in an interview: "The users of blockchain will be AI agents." She painted a vision of the future: AI will become the front-end interaction layer for all online transactions, while blockchain will take a backseat, becoming a trusted back-end infrastructure. Humans will no longer need to directly operate wallets, browse block explorers, or check transaction hashes; these complexities will be completely abstracted away by AI agents.
Almost simultaneously, the open-source AI agent project OpenClaw released version v2026.3.7-beta.1, achieving native support for GPT-5.4. This project, with over 280,000 stars on GitHub, released two major updates within two days. The official changelog included a slightly self-deprecating yet confident statement: "We fix more problems than we create—that's progress." This update not only introduced a pluggable context engine but also strengthened security mechanisms and engineering deployment capabilities—OpenClaw is evolving from an experimental agent framework into a true "agent operating system."
Meanwhile, another seemingly unrelated piece of news is circulating in the crypto community: data from RWA.xyz shows that the on-chain value of tokenized real-world assets, excluding stablecoins, has surpassed $25 billion, nearly quadrupling from approximately $6.4 billion a year ago. The on-chain size of six major asset classes, including US Treasury bonds, commodities, and private credit, has all exceeded the $1 billion threshold.
The timing of these events, occurring within the same month, is no coincidence. They all point to an emerging paradigm shift: as AI agents begin to interact autonomously with the blockchain, and as the scale of on-chain assets becomes sufficient to support an "agent economy," RWA's operating model will shift from "human management" to "AI autonomous management." This is an industrial leap that needs to be taken seriously.
I. AI is transitioning from a "co-pilot" to a "driver".
To understand the depth of this leap, we need to first see the fundamental changes that the role of AI is undergoing.
In the past few years, artificial intelligence has primarily played a "co-pilot" role in the public's perception—assisting humans in writing emails, planning trips, and generating code, but always in a passive, reactive state. Users issue commands, AI executes the commands, and the closed loop of the task is completed by humans. In this model, AI is the tool, and humans are the subject.
However, the release of the latest version of OpenClaw provides a window into this relationship, suggesting it is loosening. From March 7th to 8th, OpenClaw released two versions, 2026.3.7 and 2026.3.8, with core updates focusing on four areas: model capability upgrades, agent architecture evolution, engineering deployment optimization, and enhanced security and reliability.
Among the features that have garnered the most attention from developers is the pluggable Context Engine. This mechanism allows developers to freely mount RAG or lossless compression algorithms, solving the "forgetfulness" problem of agents in long conversations and paving the way for long-term autonomous operation. Meanwhile, ACP binding supports restart recovery, meaning that even if the server restarts, the agent can "remember" the previous communication progress and context, achieving truly persistent service.
Behind these technical details lies an important trend: AI agents are gaining "persistence" and "autonomy." They are no longer the product of one-off conversations, but digital entities that can continuously operate, learn, and perform tasks.
NEAR co-founder Polosukhin's prediction perfectly points out the application scenario of this capability: "Artificial intelligence will be at the front end, while blockchain will exist as the back end. The goal is to let your AI hide the entire blockchain—the fact that we have a block explorer is actually a failure because we haven't abstracted the technology."
In his vision, future AI agents will interact directly with blockchain protocols, autonomously completing payments, managing assets, coordinating services, and even participating in governance voting. Humans will only need to converse with the AI, telling it to "optimize my asset allocation" or "participate in voting on that proposal," and the agent will handle the rest on the blockchain.
This isn't science fiction. OpenAI's EVMbench, developed in partnership with Paradigm, is already testing the ability of AI agents to detect, patch, and exploit smart contract vulnerabilities. Circle and Stripe are racing to build stablecoin payment infrastructure for AI agents; Stripe's x402 USDC payment function on Base already supports autonomous settlement between AI agents. The "Web4.0 Marketplace" launched by the decentralized AI infrastructure protocol 0G and Alverse allows AI agents to mint and trade digital assets using cryptographic agent IDs.
An on-chain economy comprised of AI agents is moving from concept to reality.
II. From distribution to governance, every aspect of RWA is being rewritten.
When AI agents become "users" of the blockchain, the issuance, trading, management, and governance models of RWA will be systematically reshaped. This is not a localized efficiency optimization, but a paradigm shift throughout the entire lifecycle.
Asset issuance: From "manual due diligence" to "real-time verification"
Traditional RWA issuance requires the involvement of multiple human parties, including lawyers, auditors, and appraisers. Taking real estate tokenization as an example, project teams need to hire third-party appraisal agencies to issue valuation reports, law firms to conduct title investigations, and accounting firms to audit cash flow. The entire process often takes several months and is very costly.
AI agents can revolutionize this process. By connecting to data sources such as IoT devices, on-chain credit scores, and third-party APIs, AI agents can verify asset status in real time. For example, once the title deeds for a batch of goods are on-chain, and insurance policies and customs payment receipts have been verified, the AI agent can automatically trigger the tokenization process, generating corresponding RWA tokens for investors to subscribe to. The entire process is compressed from months to minutes, with human intervention minimized.
Trade Execution: From "Instruction Response" to "Strategic Game Theory"
Current RWA trading primarily relies on manual order placement or simple smart contract conditions. Investors need to switch between multiple platforms, compare prices, assess liquidity, calculate costs, and then manually execute trades. AI agents, on the other hand, can execute complex strategies. They can simultaneously monitor price differences across multiple on-chain markets and automatically execute cross-chain arbitrage; they can predict asset price trends based on macroeconomic data (such as interest rate decisions and inflation reports) and adjust their positions in advance; and they can automatically execute stop-loss or hedging operations when preset risk control thresholds are triggered. More importantly, the competition among multiple AI agents in the same market will give rise to complex dynamics that are difficult for humans to simulate—this is both a challenge and an opportunity to improve market efficiency.
Asset Management: From Monthly Reconciliation to Continuous Monitoring
Management during the life of an RWA (Recovery and Investor Asset Management) is often the most easily overlooked aspect. Routine operations such as rent collection, interest payments, collateral monitoring, and profit distribution rely on manual reconciliation and collection, which is inefficient and prone to errors. AI agents can achieve 24/7 monitoring. They can automatically allocate cash flow generated by assets to investor wallets; immediately issue margin calls when collateral value falls below a warning threshold, and even initiate liquidation procedures; and automatically handle early redemption and renewal upon maturity according to preset rules in smart contracts. For investors, this means a significant improvement in the transparency and timeliness of asset management.
Governance Participation: From Low Voter Turnout to Algorithmic Democracy
Tokenized assets typically come with governance rights, but traditional voting participation is extremely low. Most investors lack the time or willingness to delve into proposals, rendering governance a mere formality. AI proxies can change this. By analyzing proposal texts, assessing their impact on asset value, and simulating changes in returns under different voting outcomes, AI proxies can make decisions on behalf of investors. They can participate in governance continuously, rather than passively voting only at annual shareholder meetings. This makes governance a truly daily activity, rather than an occasional formality.
Third, the market is already voting with real money.
These may sound like predictions for the future, but market data is already confirming the trend. Data from RWA.xyz shows that as of March 2026, the on-chain value of tokenized real-world assets, excluding stablecoins, has exceeded $25 billion, nearly quadrupling from a year ago. Six major asset classes—US Treasury bonds, commodities, private credit, institutional alternative investment funds, corporate bonds, and non-US government debt—each have on-chain assets exceeding $1 billion. Traditional financial giants are accelerating their deployments. BlackRock launched the tokenized fund BUILD on Ethereum, Franklin Templeton migrated its US government money market fund FOBXX to the Solana blockchain, and JPMorgan Chase processed billions of dollars worth of tokenized collateralized repurchase transactions through its Kinexys platform. These institutions will not easily enter a market without prospects. The competition between Circle and Stripe in AI agent infrastructure is particularly noteworthy. These two institutions, long positioned at opposite ends of the stablecoin value chain, are penetrating each other's business areas. Circle is building its application-layer infrastructure through the Arc L1 blockchain, the CCTP cross-chain transfer protocol, and the Circle Payments Network; Stripe launched the x402 USDC payment function for AI agents on its Base platform, acquired Bridge for $1.1 billion, and is co-developing the Tempo L1 settlement chain with Paradigm. Artemis data shows that USDC on-chain transaction volume exceeded $8.4 trillion in January of this year, and the entire stablecoin market has surpassed $300 billion. This is a financial scale sufficient to support the operation of an AI agent economy. Meanwhile, OpenAI's EVMbench, developed in collaboration with Paradigm, is testing the capabilities of AI agents in smart contract security. Subsequent research shows that in EVMbench tests, the AI agent was able to detect up to 65% of real-world vulnerabilities. Although the end-to-end exploit success rate has not yet reached the level of human experts, this data is already enough to attract the attention of the security industry.
IV. Two sides of the coin: great opportunities, but also many pitfalls.
Every major technological revolution comes with both opportunities and risks, and the integration of AI agents and RWA is no exception. On the opportunity side , improved efficiency is the most direct value proposition. AI agents can operate 24/7, unaffected by human physiological limitations; they can simultaneously monitor hundreds of markets, capturing fleeting arbitrage opportunities; and they can execute complex strategies that are difficult for humans to achieve. For asset management institutions, this means the possibility of reduced operating costs and expanded management scale. New business models are also emerging. "AI agent as a service" platforms may become the next growth point: enterprises can rent professional AI agents to manage their RWA assets without building their own technical teams. New specialized agent service providers may emerge in niche areas such as cross-chain liquidity aggregation, automated market making, and algorithmic governance. Global liquidity is another promising dimension. AI agents can seamlessly access multi-chain markets, transferring assets between different blockchain networks, breaking down the liquidity barriers in the current RWA market caused by inter-chain fragmentation. When agents can freely move between different ecosystems such as Ethereum, Solana, and NEAR, the depth and breadth of the RWA market will be significantly enhanced. Challenges are equally significant . Security risks are the primary concern. AI agents, holding private keys, executing transactions, and managing assets, have become new targets for hackers. Vulnerabilities in private key management, flawed algorithm design, and adversarial attacks can all lead to asset losses. EVMbench research shows that while AI agents perform well in vulnerability detection, their success rate in real-world end-to-end exploitation scenarios is far lower than expected. This indicates that current technology is insufficient to support fully unattended asset management. Compliance challenges are equally thorny. The legal status of AI agents is unclear: if an agent's erroneous decisions lead to asset losses, who should bear the responsibility? The developer? The deployer? Or the asset holder? Regulatory attitudes differ across jurisdictions, and the global accessibility of blockchain further complicates matters. In mainland China, according to Document No. 42 jointly issued by eight departments, conducting RWA tokenization and related services within the country is illegal, and on-chain operations of AI agents must strictly adhere to this red line. Technological barriers are a real obstacle. For enterprises to embrace the AI agent economy, they need both blockchain integration capabilities and AI deployment capabilities, which presents a significant challenge for traditional businesses. Developing multi-skilled teams, selecting suitable technology partners, and designing robust governance frameworks all require time and resources.
5. Want to get on board? Do these four things first.
Faced with the emerging AI agent economy, traditional enterprises and listed companies need to formulate clear strategic paths. The first step: Asset digitization first. AI agents manage digital assets, not physical ones. Therefore, companies need to first tokenize their existing physical assets (accounts receivable, equipment, property, intellectual property, etc.) through compliant channels. For mainland Chinese companies, this means paying attention to filing channels in Hong Kong and other regions, exploring RWA's overseas expansion path within the framework permitted by Document No. 42. The second step: Pilot AI agent nodes – no need for full-scale deployment all at once. Companies can choose specific scenarios (such as cross-border payments, supply chain financing, investor relations maintenance) as pilots, cooperating with mature AI agent protocols to introduce agents for automated management. Accumulate experience from small-scale pilots, evaluate the effects, and then gradually expand the application scope. The third step: Cultivate a multi-skilled team . The AI agent economy requires a cross-disciplinary talent pool. Companies need personnel who understand blockchain technology, engineers who know how to deploy and optimize AI models, and legal experts familiar with financial compliance. Cultivating or attracting such multi-skilled talent is a guarantee of long-term competitiveness. Step 4: Participating in Standards Development. The integration of AI agents and RWA is still in its early stages; technical standards, governance rules, and compliance frameworks are all under development. Forward-thinking companies should actively participate in industry discussions and promote the development of rules that benefit their own growth.
Conclusion: The two sides of digital civilization are quietly converging.
Revisiting the two events mentioned at the beginning of this article—OpenClaw's technological breakthrough and the leap in the scale of the RWA market—they seem independent, but in fact point to the same profound historical proposition. Within the RWA Research Institute's cognitive framework, AI and blockchain have always been two sides of the same coin in digital civilization. One represents ultimate productivity, the other advanced production relations. When AI agents begin to autonomously manage on-chain assets, these two sides are undergoing an unprecedented deep integration. AI agents process information, execute strategies, and participate in games with extreme efficiency, while blockchain provides trusted asset registration, transparent rule enforcement, and trustless value transfer. This is not a simple technological superposition, but an evolution of economic organization. When assets are autonomously managed by AI agents, humans will retreat to the role of rule-makers and strategy designers. What social impact will this bring? How will governance power be distributed? How will the boundaries of responsibility be defined? These questions have no ready-made answers and require joint exploration by industry, regulators, and academia. But one thing is certain: the on-chain economy constituted by AI agents has already quietly begun in a version update in March 2026. (This article is based on publicly available information, with data as of March 12, 2026. According to Document No. 42 jointly issued by eight Chinese departments, conducting RWA tokenization and related services within mainland China is illegal. The AI proxy on-chain economy discussed in this article only applies to overseas compliance frameworks and does not constitute any investment advice.) References:
• "With over 280,000 stars, OpenClaw has undergone two major updates in two days! Adapting to GPT 5.4, say goodbye to 'gacha-style prompts'!"
• NEAR co-founder says AI agents will become major users of blockchain.
• On-chain RWA market surpasses $25 billion, nearly quadrupling in size in the past year.
• Bloomberg: Stablecoin companies bet on AI-assisted payments, but practical applications remain limited.
