Editor's Note: As AI Agents evolve from chat tools to systems that autonomously execute tasks, blockchain infrastructure is seeking a new positioning. On October 21, NEAR co-founder Illia Polosukhin and team members Alex and Anton appeared on The Rollup's "AI Super Cycle" program to discuss NEAR Intents, privacy computing, post-quantum security, and the development direction of the AI Agent economy. Compared with the competition among traditional public chains around transaction performance and ecosystem scale, NEAR is trying to integrate these technologies into a set of infrastructure serving the AI economy.

Behind this lies a more fundamental question: if future users mainly complete shopping, payments, asset trading, and other commercial activities through AI Agents, who will be responsible for connecting different markets, executing transactions, and verifying results? At the same time, as AI masters more and more personal and corporate data, how can we avoid a situation where a small number of model providers simultaneously control user information, transaction entry points, and commercial distribution mechanisms?

In the interview, Polosukhin reiterated NEAR's core judgment: AI will become the front end through which people interact with computing systems, while blockchain will be responsible for coordinating economic activity among AI, humans, and enterprises. Under this vision, NEAR Intents is no longer just a cross-chain trading tool, but may further develop into a trading network connecting financial assets, digital services, and even physical goods; NEAR AI provides privacy protection, verifiable inference, and model access capabilities.

It is worth noting that this strategy is also being combined with NEAR's token economic design. NEAR has already established a link between part of protocol revenue and token buybacks, and launched a mechanism to obtain AI computing quotas by staking NEAR. However, from the growth of cross-chain transaction revenue to the formation of widely used AI economic infrastructure, issues such as user adoption, commercial demand, and value capture still need to be addressed. What this interview reflects is more NEAR's judgment on the future market structure, rather than the full validation of this model.

The following is a translation of the original text:

AI as the front end, blockchain as the back end

Over the past few years, NEAR's development direction has not always appeared clear to the outside world. From Chain Abstraction and NEAR Intents to decentralized AI, privacy computing, and post-quantum cryptography, these businesses seem to correspond to different markets.

But in Polosukhin's view, these technologies have always revolved around the same goal.

"My core thesis has always been that AI will become the interface through which we interact with computing systems, while blockchain will coordinate the relationships between AI, humans, and enterprises."

Under this architecture, blockchain primarily handles underlying settlement and security guarantees, AI is responsible for understanding user needs and interacting with users, and NEAR Intents connects different markets to complete specific tasks.

Intents, an intent-driven transaction architecture, allows users to directly express the outcomes they want to achieve without having to handle all the technical steps required to complete a transaction themselves. For cross-chain transactions, users only need to specify the asset they wish to exchange, and the system can complete the cross-chain operation through the corresponding execution mechanism.

NEAR hopes to further expand the scope of this architecture.

Polosukhin cited examples in the developing Agent Market, where services have emerged that can analyze whether websites and products are easily recommended by other AI Agents, and Agents that can generate professional photos based on selfies uploaded by users and complete physical printing and mailing.

In the latter example, users only need to express the need "I want a professional photo" and upload a selfie, and the rest of the work can be coordinated and completed by the system.

This means that NEAR Intents may in the future process not only financial transaction instructions, but also digital service delivery, commercial tasks, and physical goods transactions.

Alex summarized this positioning as the "Market of Markets."

He believes that the market has not yet formed a unified narrative around NEAR. Some see it as the SWIFT between different blockchains, some consider it infrastructure that aggregates various markets, and others view it as the underlying network of the AI Agent economy.

Alex does not consider this fragmented positioning a bad thing. On the contrary, he believes NEAR's advantage lies in its ability to collaborate with other blockchain and application ecosystems, rather than building its development on replacing other networks.

"Our focus has always been on building correct, reliable technology, supporting more application scenarios, and combining with technologies from other ecosystems to move away from zero-sum competition."

Therefore, the long-term direction proposed by NEAR is not to migrate all assets and applications onto a single public chain, but to enable different markets to operate collaboratively through a unified execution architecture.

Whether this concept can hold up depends largely on whether AI Agents truly become an important gateway for future commercial activities.

From cross-chain transactions to enterprise finance, privacy becomes NEAR's new entry point

As more and more financial activities migrate on-chain, the traditional blockchain feature of public transparency has also begun to become an obstacle for enterprises adopting crypto infrastructure.

Polosukhin believes that over the past decade, the main activity in the crypto industry has still been concentrated in speculation and financing, and real enterprise business has not entered on-chain at scale.

However, as the regulatory environment gradually becomes clearer and liquidity improves, enterprises have begun to have the conditions to use blockchain for actual operations. The problem is that enterprises usually cannot accept that all payments, revenue, and business dealings are observed by outsiders in real time.

For listed companies, the early disclosure of certain transaction data may even involve information disclosure and compliance issues.

"Blockchain has gradually matured to the stage where it can truly be used for commercial activities, but privacy is no longer just a matter of obscuring identity through anonymous addresses. We need real privacy protection."

This is also the background for NEAR's push for Confidential Intents.

Confidential Intents is a confidential transaction execution mechanism built on top of NEAR Intents. According to NEAR's official disclosure, it executes transactions through dedicated private shards, so that information such as transaction amounts, asset portfolios, counterparties, and execution strategies does not all have to be exposed on the public chain.

On July 7 this year, NEAR announced that Confidential Intents was fully opened to developers. Officials disclosed at the time that NEAR Intents had cumulatively processed more than $1.5 billion in ZEC trading volume, and within weeks after the launch of the confidential mode, about 42% of near.com's trading volume used the feature. The above data reflect usage on specific platforms and cannot be directly inferred as privacy transaction demand across the entire market.

Compared with the model of fully disclosing all transaction information, Confidential Intents aims to protect commercially sensitive information while retaining verifiable settlement capabilities.

However, this does not mean that NEAR has achieved fully trustless anonymous transactions. Official technical documentation shows that the system relies on private shards run by permissioned validators, making trade-offs among privacy protection, auditability, and trust assumptions; the selective disclosure feature also remains a capability being rolled out in stages.

For Polosukhin, the importance of privacy also lies in the fact that AI will significantly enhance data analysis capabilities.

"AI is getting better and better at tracking everyone, and increasingly capable of discovering correlations between different consumer behaviors. You don't want all that information exposed out there for anyone to access."

He believes that Zcash drove the growth of demand for personal financial privacy, and that enterprises' need for transaction confidentiality could become the next phase of this trend.

Alex emphasized that blockchain does not necessarily have to replace banks. More importantly, when users can choose between traditional custodial services and self-custodied, on-chain private transactions, they gain more ways to manage risk.

In other words, what NEAR hopes to capture is not just existing DeFi users, but also enterprises and institutions that need commercial confidentiality capabilities while wanting to leverage on-chain settlement and cross-chain liquidity.

NEAR AI Bets on the Private Inference Market

If NEAR Intents addresses how AI Agents execute tasks, then NEAR AI focuses more on whether the models themselves can be trusted when executing those tasks.

Polosukhin summarized NEAR AI's goals with three keywords: privacy, verifiability, and security.

He believes that as AI applications penetrate deeper into daily life and enterprise operations, users are providing increasingly sensitive information to models, including financial status, medical data, legal issues, personal relationships, as well as internal corporate documents and business processes.

The more complete the context, the higher the value AI may deliver, but users' dependence on model providers will also increase accordingly. "If all enterprise documents are handed over to a single service provider in plaintext, that party may effectively have the ability to reconstruct your entire business."

NEAR AI hopes to reduce this risk through encrypted inference and runtime verification mechanisms.

According to Polosukhin, NEAR AI can provide independently checkable proof information for model runs, including hash values of prompts and responses, hardware signatures, and related execution records. The goal is to enable users to verify the model's runtime environment and execution process, rather than relying solely on promises made by the platform.

This technical approach is also related to the development of Open-Weight Models. Open weights mean that model parameters can be used or deployed externally, but does not necessarily mean that training data, training code, and the entire model development process are fully open source.

Polosukhin stated that as the performance of open-weight models improves, their low-cost advantage is beginning to manifest in certain tasks.

He specifically mentioned that NEAR used DeepSeek models combined with a customized Agent runtime framework in formal verification research, achieving performance comparable to Claude and GPT on some complex tasks, but at approximately 100 to 200 times lower cost.

Formal verification is a technique that uses mathematical methods to check whether a program or system meets specific correctness requirements, particularly applicable to protocols and software systems with high security requirements.

It should be emphasized that this cost and performance comparison comes from Polosukhin's introduction of NEAR's research results; the interview did not provide complete benchmark tests, model versions, or cost calculation methodologies, and therefore it cannot be concluded that DeepSeek has equivalent advantages in all scenarios.

He also acknowledged that on the most difficult frontier problems, top-tier models still hold the leading position.

But in his view, not all tasks require the most expensive models. For the large number of tasks that can be accomplished with lower-cost models, open-weight models already possess practical competitiveness.

On this basis, NEAR AI also plans to open confidential hosting capabilities to third-party model providers. According to the architecture described by Polosukhin, model developers can deploy models to the platform while encrypting model weights; user-submitted data will also not be directly exposed to model providers.

This means NEAR AI aims to protect the intellectual property of model developers while also protecting users' input data. Its long-term goal is not merely to operate a set of proprietary models, but to establish a marketplace that connects both model providers and inference demanders.

In terms of business model, NEAR launched a mechanism in July for obtaining AI computing credits through staking NEAR. According to official documentation, users can form periodic computing credits through staking, using relevant AI services while retaining ownership of the underlying staked assets.

However, whether the confidential inference marketplace can compete with existing cloud services on price, model quality, latency, and developer experience still requires continued observation.

How Are AI Agents Changing Internet Business Models?

Regarding the future of AI Agents, the NEAR team has also put forward a more controversial judgment: existing internet platforms blocking AI Agents from executing transactions may not be solely due to technical or security considerations, but may also involve protection of existing business models.

Polosukhin cited Amazon as an example, noting that e-commerce platforms derive revenue not only from merchandise sales but also from search advertising, sponsored promotions, and other businesses. These business models depend on users browsing and clicking on web pages, and are influenced by search rankings and ad placement.

But the purchasing logic of AI Agents may be completely different. "For an Agent, what it scans is a set of JSON data. Whether a product is ranked first or third doesn't necessarily matter."

In this scenario, if an Agent can autonomously compare prices, product features, and fulfillment capabilities based on user needs, the commercial value of traditional search promotion slots could be undermined.

Polosukhin compared this to Google's adoption of large language models. He believes that even if new interaction methods can improve user experience, they may encounter internal resistance because they disrupt existing business revenue.

Alex, meanwhile, views such resistance as a common phenomenon in technological transitions. He used the example of automobiles replacing horse-drawn carriages to illustrate that incumbent commercial interests often oppose new technologies but may not be able to prevent their adoption in the long run.

In his view, over the next five years, some companies that hold globally important positions today may lose their advantage due to failure to adapt to technological transitions.

However, this remains Alex's prediction about the industrial competitive landscape, not a confirmed market outcome. Whether AI Agents can autonomously complete payments, procurement, and delivery still involves practical issues such as platform authorization, identity verification, consumer protection, and transaction liability.

For NEAR, this shift could create new business opportunities.

If commercial activity shifts from users directly visiting websites to AI Agents purchasing services on behalf of users, the market will need new infrastructure to handle service discovery, quotation, payment, execution, and verification.

NEAR Agent Market is being developed around this direction.

Polosukhin stated that NEAR hopes to extend the capabilities of Intents from asset trading to digital work, digital goods, and physical goods, enabling Agents to directly connect with different service providers and complete task delivery.

But whether this model can become mainstream in internet commerce still depends on whether service supply, user adoption, and cross-platform collaboration can reach sufficient scale.

From User-Owned AI to Post-Quantum Security

Beyond transactions and business models, the NEAR team also views AI ownership as a critical issue for the future.

Alex points out that today's AI chatbots may capture far more user context than traditional search engines.

In the past, search platforms primarily inferred needs based on keywords entered by users. But when users rely on AI assistants over time to handle work, life, and personal decisions, model providers may gain access to more complete and continuous personal information. If this information is centrally used for ad recommendations and commercial monetization, platforms could gain stronger behavioral influence capabilities.

Alex worries that as AI business models mature, users may even find it difficult to distinguish between their own judgments and long-term algorithmic guidance. He cites this risk as an important reason for supporting User-Owned AI.

Polosukhin further argues that user-owned AI is not just a data privacy issue—it also involves economic incentives and system control. He believes one of Bitcoin's key innovations was coordinating different participants through economic incentives to form a collectively maintained public network. AI could adopt a similar model, allowing users and developers to jointly build the system while retaining control over their own data, model usage processes, and related assets.

Among these, privacy protection and verifiability are foundational requirements. For example, users should be able to inspect the code running the model, verify the execution environment, and confirm as much as possible that the system has not been modified without authorization.

But this does not mean all AI systems can immediately achieve full decentralization. Polosukhin acknowledges that collaborative training, fine-tuning, and continuous improvement of models still present numerous R&D challenges.

Therefore, NEAR is currently focusing first on private inference infrastructure, then gradually exploring broader user ownership models.

On the security front, NEAR has also incorporated post-quantum cryptography into its overall architecture.

Near One CTO Anton Astafiev said in an interview that NEAR's original account design allowed support for multiple cryptographic signature schemes, giving it flexibility when signature algorithms need to be replaced or added in the future.

NEAR's official technical roadmap released in August this year confirms that the protocol has added support for the ML-DSA post-quantum signature scheme and plans to continue optimizing related performance and wallet compatibility.

Anton emphasizes that post-quantum security should not be simply understood as quantum computers suddenly breaking all cryptography one day. The more realistic task is to give the blockchain the ability to update signature schemes, migrate assets, and recover systems when cryptographic risks emerge.

NEAR believes that its multi-cryptography account architecture can reduce the difficulty of future migration.

However, supporting post-quantum signatures does not mean that the entire protocol and ecosystem already have comprehensive post-quantum security. The official roadmap also acknowledges that wallets, cross-chain infrastructure, and other components of the protocol still need continued upgrades.

NEAR's Next Step: From Transaction Revenue Growth to Value Capture in the AI Economy

At the end of the interview, the host turned the topic to NEAR's network economic model.

He proposed a possible development path: NEAR Intents generates revenue through cross-chain transactions and other market activities, and then uses these economic activities to support the development of NEAR AI. As more AI Agents and users rely on this network, NEAR may gradually become systemically important infrastructure in the AI economy.

This is also the token economic logic that NEAR has tried to establish in recent years.

In February of this year, NEAR officially disclosed that Intents had cumulatively processed more than $13 billion in cross-chain transaction volume, while also launching a fee distribution mechanism and directing part of the revenue toward NEAR token buybacks.

By the quarterly report in July, NEAR further linked the near.com front-end business, protocol fee capture, and token buybacks, attempting to establish a relationship between network usage growth and token economic value.

This type of design means that NEAR no longer relies solely on block space demand to build its token value narrative, but also hopes to find revenue sources from cross-chain transaction execution, application distribution, and AI computing services.

However, transaction volume, protocol revenue, and token value cannot be directly equated.

Growth in cross-chain transaction scale does not necessarily mean profit grows in tandem; the fees obtained by the protocol also need to distinguish between distributor shares, actual net revenue, and the portion that can ultimately enter the token buyback mechanism.

Similarly, obtaining AI computing credits by staking NEAR may increase token usage demand or lock-up demand, but whether it can sustainably generate economic value still depends on the real usage scale of AI services and users' willingness to pay.

At the end of the interview, the host summarized this possibility as: Can NEAR gradually evolve from a story about revenue growth into a story about the currency and network value of the AI economy?

Since the provided interview transcript ends at this question and does not include further responses from the interviewee, this judgment should still be regarded as a market framework proposed by the host, rather than a conclusion confirmed by the NEAR team in this interview.

For NEAR, the more meaningful validation metrics going forward are not whether it can simultaneously tell the three narratives of cross-chain, privacy, and AI, but whether these businesses can form mutually reinforcing real demand: whether Intents continues to generate capturable net revenue, whether AI inference and Agent Market attract external developers and paying users, and how much of the new economic activity can ultimately settle into the NEAR network and token system.

Only when these connections gradually become measurable can NEAR's proposition that "AI is the front end, blockchain is the back end" potentially evolve from a technical vision into a verifiable economic model.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia