A16z 2026 AI Three Major Predictions: Rise of Research-Oriented AI, KYA Takes Over KYC, Internet Invisible Tax Crisis

As the capabilities of reasoning models rapidly evolve in the second half of 2025, the focus of AI competition in 2026 is shifting from smarter to more actionable, trustworthy, and properly valued. Three members of the a16z crypto research and investment team recently shared their insights on the three major trends in AI development for 2026, from research workflows, agent infrastructure, to the economic models of open networks.

Harvard professor Scott Kominers predicts that by 2026, AI will transition from assistants to research partners, offering creative perspectives. He states that by November 2025, he was already able to interact with models using abstract instructions similar to those given to PhD students, receiving novel and correct answers.

Circle co-founder Sean Neville believes that enabling AI agents to conduct transactions as autonomous entities—traceable, authorized, and verifiable under the Know Your Agent (KYA) framework—will be a key trend.

Liz Harkavy of the a16z crypto investment team argues that AI extracting content from the internet and expanding itself without contributing advertising traffic leads to a high mismatch between the contextual and operational layers of the web. She advocates that rewards should be distributed to all entities contributing information, data, or content when an agent successfully completes a task. She also mentions that blockchain-supported nanopayments and more mature attribution standards could be feasible technical solutions.

Trend 1: AI transitions from assistant to research partner, capable of undertaking substantive research tasks

Scott Kominers, a researcher at a16z crypto and a Harvard Business School professor, notes that early 2025, enabling consumer-grade AI models to understand his research workflows was quite challenging. However, by November 2025, he could interact with models using instructions akin to guiding a PhD student, with the models sometimes providing novel and correctly executed answers.

In 2026, a new scholarly research style will emerge

Kominers points out that AI use in research is becoming more widespread, especially in fields requiring reasoning. Models have begun directly assisting exploration and can even automatically solve difficult problems like Putnam-level math competitions. Which disciplines will benefit most and how remains an open question.

He anticipates a new polymathic research style in 2026: researchers will focus more on proposing cross-concept conjectures and quickly extrapolating verifiable directions from still largely conjectural answers.

AI research evolution still carries hallucination risks, but cryptographic techniques can help

He admits that this research approach inevitably involves inaccuracies and hallucination risks. However, when models become sufficiently intelligent, providing an abstract space for divergence may lead to breakthroughs similar to human creativity. He envisions that in 2026, research workflows will resemble agent-wrapping-agent architectures: multiple layers of models evaluating, verifying, and synthesizing conclusions.

However, Kominers also warns that large-scale deployment of such reasoning agent clusters will require better model interoperability and methods to identify and reasonably compensate each model’s contribution. He believes cryptographic technology could assist with these issues.

Trend 2: From KYC to KYA—knowing your agent as a new bottleneck in agent economy

Sean Neville, co-founder of Circle, architect of USDC, and current CEO of Catena Labs, focuses on the key bottleneck in the agent economy: shifting from intelligence to identity.

Neville points out that in fields like financial services, the number of non-human identities has far surpassed human employees, reaching a ratio of 96:1. Yet, most of these identities are ghost accounts incapable of opening accounts or bearing responsibility. Therefore, he advocates that the next primitive is KYA (Know Your Agent).

By definition, KYA aims to solve the problem that if an agent is to act on behalf of a principal, it must possess verifiable, traceable, and accountable credentials. Just as humans need credit scores to borrow, agents need cryptographically signed credentials linking their principals, behavioral constraints, and liabilities. Until KYA is established, merchants and service providers will continue to block agent access at firewalls to prevent fraud, abuse, and unclear liabilities.

He also candidly states that the decades-long KYC industry and regulatory framework may have only a few months to explore and implement KYA.

Trend 3: AI agents impose invisible taxes on the web, extracting content value and bypassing revenue

Liz Harkavy of the a16z crypto team focuses on how the economic foundation of open networks is being reshaped by agents. She describes the rise of AI agents as imposing an “invisible tax”: agents extract content from ad-supported websites (she calls this the Context layer), providing users with more convenient answers and operations (Execution layer), but systematically bypassing the revenue sources supporting content creation—such as ad impressions, subscriptions, and traffic referrals.

Harkavy believes this causes a high mismatch of interests between the web’s contextual and operational layers: content providers bear the costs, while agents and platforms absorb the value, and the original monetization pathways are cut off. She notes that current AI licensing deals are mostly superficial, often compensating content creators with a small share of lost traffic revenue, which may not be sustainable long-term.

She advocates that to prevent the open web from being hollowed out—and to protect the diverse content sources AI depends on—2026 will require large-scale deployment of technical and economic solutions: such as new sponsorship content models, micro-attribution systems, or other innovative funding mechanisms. The key is to advance static licensing to real-time, usage-based compensation, enabling value to flow automatically.

Additionally, when an agent successfully completes a task, rewards should be distributed to all contributing entities—information providers, data sources, or content creators. She also mentions that blockchain-supported nanopayments and more mature attribution standards could be feasible technical paths.

This article, “a16z 2026 AI Three Major Predictions: Rise of Research AI, KYA Replaces KYC, Web’s Invisible Tax Crisis,” first appeared on Chain News ABMedia.