Episode Summary
Executive Summary: Ilya Pusakin argues that AI and blockchain are converging: AI becomes the user interface and blockchain the trust/back-end layer for identity, payments, governance, and coordination. He says agent utility is blocked mainly by security, privacy, and context limits, and presents IronClaw and Near AI Cloud as a private, user-owned AI stack that can safely run agents and unlock more autonomous workflows.
Main Topics: AI as the new operating system/interface (Priority: 5/5): Ilya’s core thesis is that AI will replace much of the traditional app/OS layer by composing software and acting as the primary interface between users and computing systems. Blockchain as trust, coordination, and governance infrastructure (Priority: 5/5): He frames blockchain as the back-end that supplies root of trust, identity, marketplaces, upgradeability, and governance for AI-driven systems. Why agents are not yet broadly useful (Priority: 5/5): The discussion centers on why AI agents still feel clunky and unproductive: limited context windows, poor judgment, lack of safe access to secrets, and inadequate workflow coordination. IronClaw: secure, policy-gated agent execution (Priority: 5/5): IronClaw is presented as a defense-in-depth system that vaults credentials, isolates tools, detects prompt injection/exfiltration, and prevents agents from misusing private data or capabilities. Private/confidential AI and user-owned inference (Priority: 4/5): Near AI Cloud offers confidential inference using open-weight models, secure enclaves, attestation, and MPC so neither the provider nor hardware host can access user data. Agent marketplaces and autonomous businesses (Priority: 4/5): Ilya envisions AI agents participating in markets, competitions, and potentially autonomous businesses governed by tokens and smart-contract-based rules. Accelerationism, human values, and sovereignty (Priority: 4/5): The conversation closes on the social implications of AI: a push for empowering individuals, preserving privacy, and preventing centralized misuse while still accelerating capability.
Key Arguments: AI will increasingly function as the operating system of personal computing, composing software and handling tasks on behalf of users rather than merely answering questions. Blockchain is useful not as an alternative society, but as the trust layer that coordinates identities, upgrades, markets, money, and governance for both humans and AIs. Agent utility is constrained less by raw intelligence than by lack of trust, context, and safe access to sensitive information. Current agent tools can become much more useful once secrets are vaulted, actions are policy-checked, and tool execution is sandboxed. Privacy is a major blocker because many AI workflows currently send secrets to centralized model providers and even route data through third-party startups. Confidential AI infrastructure (TEEs, MPC, attestations, open-weight models) can make AI inference user-owned and inspectable without exposing prompts or logs. Longer context windows and memory systems are essential because agents need human-like continuity to perform multi-step work reliably. As AI increases execution capacity, organizations may need to shift from hierarchy toward internal marketplaces and competition-based task allocation. Autonomous agents are likely to emerge, but the safer near-term form is autonomous businesses with explicit missions, governance, and constraints. Crypto and AI culture can be bridged through products that prove real utility: private inference, agent marketplaces, and governance systems for AI. For builders, the advantage increasingly comes from asking the right questions, defining success criteria, and creating network effects rather than pure execution alone.
Data Points: Transformer paper co-authors: 1 of 8 - Ilya says he was one of the eight co-authors of Attention Is All You Need. Year Near AI was founded: 2017 - He says he left Google in 2017 to co-found Near AI. Data-labeling pay per task: 15 cents - He describes paying crowdworkers around the world 15 cents per coding task to generate training data. Timeline from paper to ChatGPT moment: about 5 years - He notes the public AI breakthrough arrived around 2022 after the 2017 paper. NeoCrowd employees: 0 employees - He cites NeoCrowd as an example of a blockchain-enabled crowdsourcing system running without employees. NeoCrowd workers: thousands - He says the system employed thousands of people worldwide. Marketplace agents: about 500-600 agents - He says their agent marketplace currently has roughly 500 to 600 agents. TEE overhead: ~1-5% - He claims confidential computing overhead is usually low, around 1% to 5% depending on model and networking. Galaxy platform assets: over $12 billion - A sponsor segment states Galaxy has over $12B in assets on the platform. Galaxy loan book: $1.8 billion average in late 2025 - Sponsor segment cites an average $1.8B loan book in late 2025. Galaxy Helios power capacity: 1.6 gigawatts - Sponsor segment says the Helios data center campus has more than 1.6 GW of approved power capacity. OpenClaw memory/context issue: million-token context referenced - Ilya and the hosts discuss million-token context as a current frontier and bottleneck for agents.
Pivotal Quotes: "AI is a user interface, blockchain is a back-end." — Ilya Pusakin: He summarizes his thesis on how the two technologies fit together. "The keys never touch LLM." — Ilya Pusakin: He explains IronClaw’s core security property for protecting credentials from model exposure. "The AI needs to be on your side because, yeah, if this is the only way you actually perceive reality..." — Ilya Pusakin: He argues for user-owned AI and warns about centralized control of model behavior.
Implications: The piece suggests the next AI wave depends on confidential, user-controlled infrastructure. If privacy, trust, and context improve, agents could move from gimmicks to real work, reshape organizations, and create new crypto-native markets and governance systems.