Author: a16z crypto

Translation: TechFlow

This Year, AI Will Take On More Substantive Research Tasks

As a mathematical economist, back in January 2025 it was difficult for me to get consumer-grade AI models to understand my workflow. But by November 2025, I could already issue abstract instructions to AI models much like I would to a PhD student—and sometimes they returned novel and correct answers. Beyond my personal experience, AI is being more broadly applied in research, especially in reasoning tasks. These models not only directly assist in the discovery process but can also independently solve challenging problems such as Putnam questions (perhaps the hardest undergraduate math competition in the world).

It remains unclear exactly which fields will benefit most from this kind of research assistance, or precisely how it will be implemented. But I expect that this year, AI research will drive and reward an entirely new “generalist” research style—one that emphasizes conceptualizing relationships among ideas and rapidly drawing inferences from answers that may still be somewhat speculative.

These answers might not be fully accurate, yet they can still guide research in the right direction (at least within a certain topological framework). Ironically, this resembles harnessing the power of model "hallucinations": when models are "smart enough," giving them abstract space to explore ideas may still produce nonsensical results—but occasionally lead to breakthrough discoveries, just as humans can be most creative when working non-linearly or without clear direction.

Reasoning in this way requires a new style of AI workflow—not just simple “agent-to-agent” interactions, but complex “agent-nested-within-agent” collaboration. In this mode, models at different levels help researchers evaluate early-stage proposals and progressively distill insights. I’ve already used this approach to write academic papers, while others have applied it to patent searches, creating new forms of art, and even—unfortunately—discovering new smart contract attack vectors.

However, operating these nested reasoning agents for research still requires better interoperability between models, and a method to identify and appropriately compensate each model’s contribution—challenges where blockchain technology may offer solutions.

—Scott Kominers (@skominers), Member of a16z Crypto Research Team, Professor at Harvard Business School

From KYC to KYA: The Shift in Identity Verification

The bottleneck in the agent economy is shifting from intelligence to identity verification. In financial services, the number of “non-human identities” now exceeds human employees by 96 times—yet these “identities” remain unbanked “ghosts.”

The missing piece of infrastructure is “Know Your Agent” (KYA). Just as humans need credit scores to obtain loans, agents require cryptographically signed credentials to conduct transactions—credentials that link agents to their principals, constraints, and responsibilities. Until this infrastructure exists, businesses will continue blocking agents at their firewalls.

The industry that spent decades building KYC (Know Your Customer) systems now has only months to figure out how to implement KYA.

—Sean Neville (@psneville), Co-founder of Circle, Architect of USDC; CEO of Catena Labs

Solving the 'Invisible Tax' on Open Networks: An Economic Challenge for the AI Era

The rise of AI agents is imposing an “invisible tax” on open networks, fundamentally disrupting their economic foundations. This disruption stems from a growing mismatch between the internet’s “context layer” and its “execution layer”: currently, AI agents extract data from ad-supported websites (the context layer), providing convenience to users while systematically bypassing the revenue mechanisms—such as ads and subscriptions—that sustain content creation.

To prevent the gradual erosion of the open web (and protect the diverse content fueling AI), we need large-scale technological and economic solutions. These could include next-generation sponsored content models, micro-attribution systems, or other novel funding mechanisms. However, existing AI licensing agreements have proven financially unsustainable, often compensating content providers for only a small fraction of the revenue lost due to AI-driven traffic diversion.

The web urgently needs a new techno-economic model enabling automatic value flow. A key shift over the coming year will be moving from static licensing models toward compensation mechanisms based on real-time usage. This means testing and scaling systems—potentially using blockchain-enabled nanopayments and sophisticated attribution standards—to automatically reward every entity that contributes information enabling an AI agent to successfully complete a task.

—Liz Harkavy (@liz_harkavy), a16z Crypto Investment Team