Illia Polosukhin is one of the eight co-authors of the 2017 Google Brain paper "Attention Is All You Need", the research that introduced the Transformer architecture and made every modern large language model possible. He is also the co-founder and CEO of NEAR Protocol, a top-30 Layer-1 blockchain currently positioning itself as the settlement layer for AI agents that transact onchain.
That combination is rare enough to matter. There are crypto founders who talk about AI, and there are AI researchers who occasionally tweet about crypto, but the overlap between "co-wrote the foundational LLM paper" and "runs a multi-billion-dollar L1" is a set of one. In March 2026 Polosukhin told CoinDesk that "the users of blockchain will be AI agents, not humans," and the line stuck because of who said it.
From Google Brain to the Transformer Paper
Polosukhin grew up in Kharkiv, Ukraine, and studied applied mathematics before joining Google in 2014 as a software engineer on the search team. He moved to Google Brain shortly after, working on natural language understanding when the field was still dominated by recurrent neural networks and long short-term memory architectures that struggled with long-range dependencies.
The June 2017 paper changed that. Polosukhin and seven co-authors, including Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan Gomez, and Lukasz Kaiser, proposed an architecture that dropped recurrence entirely and relied only on self-attention to process sequences in parallel. The model trained faster, generalized better, and scaled in a way that earlier architectures could not. Every major LLM that came after, including GPT, Claude, Gemini, and Llama, is built on the Transformer or one of its direct descendants.
Polosukhin's specific contribution was on the machine-translation side, where the Transformer first proved its value before generative pretraining made it the default architecture for almost every language task. He left Google in 2017 the same year the paper was published, which is the part most retail crypto traders miss. He didn't drift from AI into crypto looking for a new lane. He walked out of the lab that just changed AI forever and started a blockchain company.
What NEAR Actually Is and Why the AI Angle Was Always There
Polosukhin co-founded NEAR in 2017 with Alexander Skidanov, a former MemSQL engineer, after a brief detour building NEAR.ai, a tool that aimed to teach machines to write code from natural-language specs. The team hit a wall on the AI side around scaling payments and coordination between human contributors, which is how they ended up building a blockchain in the first place. NEAR Protocol launched its mainnet in April 2020 with a sharded proof-of-stake design called Nightshade, human-readable account names instead of hex addresses, and account abstraction baked in at the protocol level rather than bolted on as an afterthought.
Those choices look different in 2026 than they did in 2020. Account abstraction means a wallet can be programmed to make decisions, pay gas in any token, batch transactions, and act on behalf of a user without that user signing every step. For a human, that is a quality-of-life feature. For an AI agent that needs to execute thousands of micro-transactions across multiple chains without human input, it is the only viable execution model.
NEAR also runs Aurora, an EVM-compatible runtime that lets Ethereum contracts deploy without rewriting, and a chain abstraction stack called Chain Signatures that lets a NEAR account directly sign transactions on Bitcoin, Ethereum, and other chains using multi-party computation. The technical pieces were assembled for general scaling and ease of use, but they happen to be exactly what an autonomous agent needs to operate across the broader onchain economy.
The March 2026 "AI Agents Will Be the Users" Call
Polosukhin's argument, laid out in his CoinDesk interview and at the EthCC stage earlier this year, is that thinking about blockchain UX for humans is a category error. Humans were never going to type addresses into wallets, hold their own seed phrases, and approve sixteen transactions in sequence at scale. Agents will. And the chain that wins is the one designed for an agent's constraints, not a human's expectations.
The thesis has three concrete pieces. First, transactions need to be cheap and fast enough that an agent can execute hundreds per minute without burning a treasury, which is why NEAR's average finality runs near one second at sub-cent fees. Second, accounts need to be programmable enough that an agent can hold limited authority over a user's funds without exposing the master key, which is the account-abstraction story. Third, the chain needs to reach across other chains natively, because no real economic activity is going to confine itself to a single L1, which is the chain-signatures story.
Davide Crapis at the Ethereum Foundation has been pushing a parallel narrative under the dAI banner, focused on Ethereum's role in agent commerce, and Jesse Pollak at Coinbase built x402 on Base to handle stablecoin payments between agents. Polosukhin's version of the thesis is older, dates back to NEAR's original 2017 framing, and was the explicit reason he left Google in the first place. The market is now catching up to a position he has held publicly for almost a decade.
What NEAR Is Shipping to Back the Thesis
The roadmap items are concrete. NEAR Intents, a system that lets users or agents declare what they want to happen rather than how to execute it, went live in late 2025 and now routes cross-chain swaps through a network of solvers who compete to fill orders. NEAR AI, an open-source research project led by Polosukhin directly, is building agent infrastructure that runs inference and identity verification with privacy guarantees the team calls Private ML.
Chain Signatures, the MPC-based primitive that lets a NEAR account control assets on other chains, has been live since 2024 and is being used by wallets and agent frameworks to execute Bitcoin and Ethereum transactions without bridges. The token economics matter too. NEAR is currently the gas and staking asset for a network that processes more than a million transactions a day, with most of that activity coming from non-financial use cases like meta-transactions, account creation, and intent submissions rather than DeFi degens.
Token holders are watching three things. Active agent transactions as a share of total network activity, which the team reports in its quarterly transparency updates. The roster of AI projects choosing NEAR for inference and payment rails over Ethereum or Solana. And if NEAR can keep its sub-second, sub-cent execution profile as agent traffic scales, because the moment fees spike or finality slips, agents will route around the chain the same way they route around any other bottleneck.
What to Watch Next
The honest read is that Polosukhin's bet is unprovable in 2026 the same way the Transformer paper's full impact was unprovable in 2017. Agent commerce is real but small. Most onchain volume is still humans speculating on tokens, and the share of activity that is actually autonomous agents transacting with other agents is in the low single digits across every chain that bothers to measure it.
What changes the picture is adoption curves. If agent transactions grow at the rate consumer LLM usage grew between 2022 and 2025, the chain that owns the rails for that traffic captures meaningful value. If they don't, NEAR is a competently engineered L1 that solved problems the market didn't end up caring about at scale. Polosukhin has been clear that he is willing to be wrong publicly for a decade if the technical work compounds in the right direction. That track record, plus the AI credentials, is the case for paying attention to what NEAR ships over the next 12 to 18 months.
Frequently Asked Questions
Did Illia Polosukhin actually co-author the Transformer paper?
Yes. He is listed as the eighth and final author on "Attention Is All You Need," published in June 2017 by eight Google researchers, and the paper is the foundational reference for every modern large language model. His contribution centered on the machine-translation experiments that first demonstrated the architecture's value before it was generalized to other language tasks.
How is NEAR Protocol different from Ethereum or Solana for AI agents?
NEAR uses sharded proof-of-stake to deliver around one-second finality at sub-cent fees, has account abstraction built into the base protocol rather than added through wallet smart contracts, and offers Chain Signatures so a NEAR account can directly sign transactions on Bitcoin, Ethereum, and other chains. Ethereum and Solana both have strong agent ecosystems, but NEAR's specific combination of native account abstraction and chain abstraction was designed earlier and more deliberately around the constraint that agents, not humans, would do most of the signing.
What is Polosukhin's role at NEAR in 2026?
He is the CEO of the NEAR Foundation and the public-facing technical voice of the protocol. Alexander Skidanov, his co-founder, leads core protocol research, and Polosukhin focuses on ecosystem direction, AI integration through the NEAR AI project, and external communication including conference keynotes and media interviews.
Has Polosukhin sold his NEAR tokens?
Public onchain data does not show any large outbound transfers from his disclosed addresses, and team token unlocks vested over four years from mainnet launch in April 2020. The NEAR Foundation publishes treasury and team-allocation transparency reports quarterly, and Polosukhin has stated publicly that he holds the majority of his original allocation, though no founder allocation is locked indefinitely once vesting is complete.
Bottom Line
Polosukhin is the rare founder who has top-tier credentials in both fields his thesis depends on, which is why the AI-agents-as-blockchain-users argument carries more weight when he says it than when a generalist VC says the same thing. The next 12 months are the test. Watch for agent transaction share on NEAR breaking above five percent of total activity, watch for a major AI lab choosing NEAR rails for payments or identity, and watch the NEAR Intents and Chain Signatures usage numbers in the foundation's quarterly reports. If those three lines move up together, the thesis is starting to validate. If they flatline through year-end, NEAR remains a strong L1 looking for the use case it was always designed to serve.
This article is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency trading involves substantial risk. Always conduct your own research before making trading decisions.
