Author: @BlazingKevin_ , the Researcher at Movemaker
Introduction: The Structural Leap from Generative AI to "Agent Behavior"
In 2026, the field of artificial intelligence will undergo a structural leap from "generative capabilities" to "Agent-based action capabilities." If 2023-2024 was about the amazing language generation capabilities of large language models, then 2026 will mark the formal establishment of the "AI Agent Economy."
Based on the predictions and analysis of the a16z Crypto research team, our further research found that 2026 will be a year of deep integration between AI, a productivity tool, and Crypto, a value distribution layer.
AI is no longer just a passive tool responding to human commands, but an active participant with the ability to reason, plan, trade, and discover autonomously.
Based on a16z Crypto's outlook report, the three core trends that will reshape the AI+Crypto landscape in 2026 are:
- A new paradigm for scientific research : from single agents to "Agent-Wrapping-Agent".
- The financial infrastructure revolution : from KYC to KYA (Know Your Agent).
- Economic Model Restructuring : Addressing the "Hidden Tax" Crisis Facing Open Networks Through Payments and Programmable IP.
These three trends are not isolated: the shift in scientific research paradigms depends on advanced collaboration between agents; advanced collaboration requires agents to have verifiable identities (KYA); and agents with identities must follow new value exchange protocols when acquiring data.
1. The New Era of Polymaths: The "Agent-Wrapping-Agent" Architecture in Advanced Research
Starting this year, the definition of "AI-assisted research" will undergo a qualitative leap.
We are no longer talking about simple literature retrieval or text summarization, but witnessing AI systems capable of substantive reasoning, hypothesis generation, and even autonomously solving doctoral-level problems.
The core driving force behind this transformation lies in shifting from linear hinting engineering based on a single model to complex, recursive AWA workflows.
1.1 Breakthrough in Reasoning Ability: Crossing the Boundaries of Pattern Matching
Scott Kominers of a16z points out that AI models are evolving from simply understanding instructions to being able to receive abstract instructions (like guiding a doctoral student) and return novel and correctly executed answers. Recent technological advancements indicate that AI models are breaking through the "random parrot" ceiling, exhibiting slow, deliberate reasoning abilities similar to human "systematic" thinking.
1.1.1 “Useful Illusion”
As reasoning abilities improve, a new style of "polymath" research is emerging. Scott describes this style as: "Using AI to cross disciplinary boundaries and speculate on possible deep connections between topology and economics, biology and materials science."
The "illusionary" nature of large models, which has been criticized, is being reconstructed into a "generative exploration" mechanism within the context of scientific discovery:
- Protein design case study : Researchers at the University of Washington used the "family illusion" concept to generate over 1 million unique protein structures that do not exist in nature. Among them, the novel luciferase screened out has catalytic activity comparable to natural enzymes, but with higher substrate specificity.
- Fluid dynamics discovery : Using physical information neural networks (PINNs), researchers have discovered new unstable singularities in the Navier-Stokes equations, which reveal previously unknown patterns in fluid motion.
The core of this research style lies in allowing the model to "think wildly" in the abstract space to generate high-entropy conjectures, and then using a rigorous logical verifier to filter the conjectures.
1.2 Detailed Explanation of AWA Architecture
To harness this powerful reasoning and generative capability, research workflows are shifting from a flattened to a hierarchical structure. AWA ( Automatic Web Application) refers not only to dialogue between multiple agents, but also to a recursive, hierarchical control structure.
1.2.1 Orchestrator-Executor Pattern
This is currently the most mainstream AWA implementation model. A "principal researcher" agent is responsible for maintaining the global context and research objectives, and decomposing and distributing tasks to a dedicated group of "executor" agents.
Architectural advantages : Anthropic data shows that a multi-agent system consisting of Claude Opus as the master agent and Claude Sonnet as the sub-agent outperforms a single Claude Opus agent in complex research tasks by 90.2% .
This performance improvement is primarily due to context isolation—the dominant agent does not need to process redundant information for each subtask, thus maintaining the clarity of reasoning.
1.2.2 Recursive Self-Improvement and the MOSAIC Framework
Another key feature of the AWA architecture is the introduction of a reflexion loop. When the underlying agent fails to execute a task, the error message is fed back to a "critic" agent for analysis and correction.
The MOSAIC framework (Multi-Agent System for AI-driven Code generation) significantly improves the accuracy of scientific code generation without relying on verification test cases by introducing specialized "self-reflective agents" and "principle generation agents." This closed loop of "trial and error-reflection-retry" simulates the thought process of human scientists when faced with experimental failures.
1.3 Case Study: Sakana AI's "AI Scientist"
The most compelling AWA application case in 2025 is Sakana AI's "The AI Scientist" system. This is a system designed to automate the entire lifecycle of scientific discovery.
1.3.1 Fully Automated Scientific Research Closed-Loop Process
- Idea generation : Based on a starting code template (such as NanoGPT), the system uses LLM as a "mutation operator" to brainstorm diverse research directions and calls the Semantic Scholar API to retrieve literature to ensure novelty.
- Experimental Iteration : The "Experimenter" Agent writes and executes code. If the experiment fails, the system will capture error logs using the Aider tool and autonomously repair the code until a visual chart is obtained.
- Paper writing : The "Writer" Agent uses LaTeX to write complete scientific papers, including abstracts, methods, experimental results, and automatically finds and generates BibTeX files by citing references.
- Automated peer review : The generated paper is submitted to a simulated "reviewer" agent, which scores it according to the standards of top conferences (such as NeurIPS). The system can even make multiple revisions based on the reviewers' comments.
1.3.2 Economic Benefits and Quality
The economic efficiency of the "AI scientist" system is astonishing: the computational cost of generating a complete research paper is only about $15 . The paper generated by this system, "Compositional Regularization," even successfully passed peer review at the ICLR workshop. While limitations such as citation illusions and logical flaws still exist, this case demonstrates that AI already possesses the ability not only to assist research but also to execute the entire research process.
2. Identity-related commands: From KYC to KYA
As agents are empowered to perform tasks and transactions, the digital economy faces an unprecedented identity crisis. Sean Neville (CEO of Catena Labs) warns that the number of "non-human identities" in the financial services sector is 96 times the number of human employees, and in some statistics, it's as high as 100:1. These agents—unbanked, unverified, yet operating at machine speed—represent a massive compliance black hole. The industry is urgently shifting from traditional KYC to KYA (Know Your Agent) .
2.1 The Emergence and Risks of Non-Human Identity (NHI)
2.1.1 “Shadow AI” and the Imbalance of 96:1
45% of financial services institutions acknowledge the existence of unapproved "shadow AI agents" within their organizations. These agents create "identity silos" outside of the formal governance framework.
- Risk scenarios : A test agent used for cloud resource optimization may autonomously purchase expensive reserved instances without human intervention; or a trading bot may trigger erroneous sell orders during market volatility.
- The attribution dilemma : When an agent violates regulations, who is responsible? The engineer who developed it? The manager who deployed it? Or the vendor that provided the underlying model? Without KYA (Know Your Customer), these responsibilities cannot be determined.
2.2 KYA Framework: The Cornerstone of Trust in the Machine Economy
KYA is not just about issuing ID cards, but about establishing a complete digital identity system that includes the subject, credentials, permissions, and reputation.
2.2.1 The Three Pillars of KYA
- Subject : The entity legally responsible to the Agent. The agent must be encrypted and linked to a KYC/KYB verified human or business account.
- Agent Identity : A unique digital identity based on a decentralized identifier (DID ). DIDs are cryptographically generated, immutable, and portable across platforms.
- Mandate/Delegation : A statement of authorization issued through verifiable credentials (VCs). For example, a VC might state: "This Agent is authorized to make purchases on behalf of Alice on Amazon, up to a maximum of $500."
2.2.2 Cryptographic Bindings and Trust Chains
When an Agent initiates a transaction, it issues a VC (Transaction Value). The verifier does not need to trust the Agent itself; it only needs to verify that the digital signature on the VC comes from a trusted issuer. This mechanism creates a "chain of trust": the bank trusts the company -> the company issues the VC to the Agent -> the merchant verifies the VC -> the transaction is approved.
2.3 The Protocol Stack Debate: Standardizing Agent Identity
2.3.1 Skyfire and KYAPay Protocol
Skyfire launched the KYAPay open standard, whose core innovation lies in composite tokens:
- kya token : Contains identity information (such as "verified enterprise agent").
- Pay token : Contains payment capability (e.g., "pre-authorized 10 USDC").
- kya+pay token : Packages identity with payment, allowing agents to complete "visitor checkout" without manual form filling.
2.3.2 Catena Labs and ACK (Agent Commerce Kit)
Catena Labs, founded by USDC architect Sean, has launched ACK , aiming to create "HTTP for smart agent commerce." ACK emphasizes leveraging the W3C DID standard and account abstraction to allow agents to directly control on-chain smart contract wallets, achieving stronger security than API keys.
2.3.3 Google AP2 and x402 Extensions
Google's Agent Payments Protocol (AP2) uses "authorization letters" to manage permissions and has partnered with Coinbase to develop the AP2 x402 extension , which integrates crypto payment standards directly into the protocol.
2.4 Agent Credit Scoring and Risk Control
KYA is also the beginning of a reputation system.
- On-chain reputation (ERC-7007) : Through ERC-7007 (Verifiable AI-generated content token standard), every successful interaction of the Agent (such as timely payment, generation of high-quality code) can be recorded on the chain, forming a verifiable history.
- Real-time circuit breaker : Financial institutions are deploying AI gateways that can immediately revoke a transaction agent's trading activity (VC) if the agent's behavior deviates from the benchmark (such as high-frequency abnormal trading), triggering a "digital suppression".
3. Economic Restructuring: Addressing the "Hidden Taxes" of Open Networks
Liz of a16z points out that AI agents are levying a "hidden tax" on the open web: in order to serve users, agents massively extract data (context layer) from content websites, but systematically bypass the advertising and subscription models that support the production of this content. If this parasitic relationship is not resolved, it will lead to the depletion of the content ecosystem.
3.1 “Great Decoupling”: The Full Arrival of a Zero-Click Economy
In 2025, the digital publishing industry witnessed a “great decoupling”: search volume rose, but the number of clicks flowing to websites plummeted.
3.1.1 The brutal data of traffic erosion
- Zero-click rates are soaring : a16z predicts that traditional search engine traffic will decline by 25% by 2026. Similarweb data shows that the zero-click search rate has already risen to 65% by 2025.
- Click-through rate (CTR) collapses : DMG Media reports that when AI overviews appear above search results, the click-through rate of the content plummets by 89% . Even the top-ranked search result loses 34.5% of its clicks when presented with an AI summary.
3.2 Moving Away from Static Licensing: A New Pay-as-you-go Model
In response to this crisis, the industry is shifting from static annual data licenses (such as the deal between Reddit and OpenAI) to usage-based compensation.
3.2.1 Comet Plus Model of Perplexity
Perplexity AI's Comet Plus program is a typical early attempt:
- Mechanism : An initial revenue pool of $42.5 million is established. Revenue sharing is triggered when the AI Agent cites publisher content in its answers or visits a page on behalf of a user.
- Revenue sharing : Publishers can receive up to 80% of the relevant revenue pool. This acknowledges the commercial value of "machine access".
3.3 Technical Standards: Nanopayments and Microattribution
To extend compensation across the entire network, a series of open technical standards are being implemented.
3.3.1 Payment and x402 Protocol
The HTTP 402 status code has finally been activated. The x402 protocol establishes the standard for "machine-native payments".
- Workflow : Agent requests resources -> Server returns 402 Payment Required and price (e.g., 0.001 USDC) -> Agent automatically signs payment via L2 blockchain (e.g., Base, Solana) or Lightning Network -> Server verifies and releases data.
- Economic efficiency : Traditional payment gateways cannot handle transactions worth a few cents, while the x402, combined with a low-fee chain, reduces costs to negligible levels, making payment possible.
3.3.2 Machine-readable rights: TDMRep and C2PA
- TDMRep (Text Data Mining Reserved Protocol) : A W3C community standard that allows websites to declare in robots.txt or HTTP headers: "TDM rights reserved, pay/license required". This provides a clear binary signal for the agent.
- C2PA (Content Source and Authenticity Alliance) : By embedding tamper-proof "content credentials," it proves the original source of content. Even if content is ingested by AI, the cryptographic signature provided by C2PA ensures that the attribution chain remains unbroken, providing a basis for royalty distribution.
3.4 On-chain IP Ownership: Story Protocol
A more radical change is to tokenize intellectual property itself. Story Protocol is committed to building a "programmable IP" layer.
- Mechanism : Creators register their works as "IP assets" on Story Network.
- Automated Licensing : Assets come with a "programmable IP license." When an AI agent uses the data, the smart contract automatically executes the license terms (such as "commercial use requires a 5% royalty") and automatically distributes the revenue. This creates a highly liquid IP market without the need for lawyers.
3.5 Outlook: From SEO to AEO
By 2026, the marketing focus will shift from SEO to AEO or GEO .
- Goal : Instead of pursuing "first place in search rankings", the goal is to be **"cited" by AI** or become the "preferred data source" in its reasoning process.
- Sponsored Context : The future of advertising will be "contextual injection." Brand bidding will enter the reasoning chain of the agent; for example, having a travel agent "recall" a particular hotel when planning a trip is the best option.
4. Conclusion
The technological landscape of 2026 clearly shows that the friction between human-centric internet infrastructure and machine-centric demands is forcing a complete overhaul of the digital world.
- Research Paradigm : AI is moving from assistance to autonomy. The AWA architecture enables AI to mass-produce scientific discoveries at low cost, turning "illusion" into creativity.
- Identity System : KYA is becoming a new frontier in financial compliance, giving billions of AI agents a legitimate economic identity, enabling them to safely navigate value networks.
- Economic Model : The network economy is shifting from an attention-based advertising model to a value-based payment and programmable IP model. x402, TDMRep, and Story Protocol form the railroad tracks of the new economy, solving the problem of "hidden taxes" and ensuring that data producers remain profitable in the zero-click era.
We are witnessing the birth of the Agent Economy —an economy in which software not only helps us work, but is itself a producer, consumer, and trader.
