Illia Polosukhin On The Future Of AI, Crypto, And How Near Is Going To Win

Illia Polosukhin’s central thesis is that AI agents will need markets infrastructure to find counterparties, transact and resolve disputes, with agents serving as the interface and blockchain providing the financial and settlement layer. He sees digital work as an early use case because its results are easier to verify, and argues business-to-business agent commerce could be larger than consumer shopping. His estimates put the broader opportunity at $20 trillion to $50 trillion a year, with $20 trillion to $40 trillion of activity potentially moving on-chain. He sees NEAR’s opportunity in building an integrated stack for that future, not just a blockchain.

Polosukhin’s interest in AI began with science fiction; by age 14, he thought computers should automate coding. At Google, he worked on question answering and language understanding, and saw transformers’ parallel attention approach as an advance over sequential recurrent neural networks: it made better use of GPUs and enabled larger models, more training data and longer contexts. In 2017, he and co-founder Alex imagined people describing what they wanted computers to build rather than navigating many apps and SaaS products. Their effort to train code-generation models led them to recruit contributors across borders, but payment difficulties—and fees that could exceed contributor pay—pushed them toward crypto and the idea of a more usable network.

NEAR was designed around low-cost transactions, sharding, account abstraction and WebAssembly, with the goal of making blockchain easier for users and developers. The team underestimated the engineering effort: a planned January launch after starting in August took until October 2020, and sharding raised hard questions about routing data and maintaining availability in a permissionless system. User onboarding was another obstacle. Interviews with roughly 300 crypto companies found processes that could involve 92 steps and take 12 hours, alongside a claimed 99.99% drop-off rate. NEAR responded with smoother onboarding and meta-transactions that allow applications to spare users from handling gas themselves.

NEAR’s chain abstraction and NEAR Intents extend that usability goal: instead of choosing each transaction, network or bridge, users specify an outcome and solvers find a route to deliver it. Polosukhin sees the same model fitting AI agents, which could express needs and act across services. NEAR Intents is used for trades, swaps and bridging, including through Trust Wallet, Ledger Live and MetaMask. Polosukhin said it processes about $4 billion per month overall, with NEAR.com accounting for roughly 10% to 20%, depending on the week. NEAR.com, launched at NEARCON in February, is intended to bring assets and functions across ecosystems into one consumer product, without gas top-ups or manual network switching.

Polosukhin says fees vary by asset and collected fees are converted into NEAR, creating buying pressure; fees from perps also fund NEAR buybacks, though he gave no total. NEAR.com offers payments, swaps, stocks, perps and earning, and supports private transactions through a NEAR private shard running in secure enclaves. The company plans a financial agent that can turn a user’s strategy into transactions, but says it will operate within user-defined policies and require explicit approval to go beyond them, rather than receive unrestricted access to a private key. The policy system is also available for other products to use.

For agents to act safely, Polosukhin argues, they need evidence about what is running behind an AI endpoint. An agent using an unknown service may receive output from an unexpected or reduced model, fabricated results or malicious instructions that could compromise its machine and steal keys; safeguards from a model provider do not secure an agent configured to use an unknown endpoint. NEAR AI, he says, can provide provenance for the GPU, software and model hashes, and signatures for work routed across nodes. He argues blockchain can combine verifiability and privacy, decentralize GPU access and make compute markets more liquid than bilateral negotiations. Venice uses NEAR for its end-to-end encrypted mode, and Brave offers it as an option for its agent.

NEAR’s growth strategy is sales-first: Polosukhin says the team seeks customers or design partners before building, with partnerships spanning crypto and companies outside crypto and AI. Its systems also face adversarial use and technical risk. NEAR Shield scans NEAR and other chains for outliers and suspicious activity and can pause flagged liquidity flows; after the Bitget hack, Polosukhin said, North Korea tried to move money through NEAR Intents. A BSC bridge link was exploited for $3.8 million through repeated rounding, but the team paused and patched the system, restored it within an hour and recovered all funds. He said the link came from an ecosystem partner and had not been tested sufficiently. NEAR plans more monitoring and formal verification across its contracts; Polosukhin said its verification system ranked first on multiple benchmarks and cut verification costs by 100 times.