Virtuals Says AI Agents Are Moving Toward On-Chain Economic Roles

Virtuals co-founder Jansen Teng said the company is expanding its vision for AI agents from gaming-focused software into what it describes as an “agent society,” where autonomous systems can participate in economic activity. According to CoinDesk, Virtuals began with autonomous agents for games before moving into crypto influencers, trading agents and other agent-based software systems.

The development matters because Virtuals is positioning AI agents as potential economic participants rather than simple chat interfaces. Teng said the next phase could include agents that control wallets, trade with one another and perform specialized tasks, allowing them to earn, spend and coordinate with less constant human direction.

Virtuals is organizing its work around five areas: digital agents, physical agents and robots, coordination between agents, capital formation and governance systems. Teng described the longer-term goal as a “parallel society” in which agents operate in a permissionless economy and collaborate at scale.

The model also introduces risks. Teng pointed to incorrect user intent, service failures and scams as key failure points, and said Virtuals is working on intent verification, escrow-based transaction standards and reputation frameworks. He also argued that staking and reputation could eventually influence how much trust or capital an agent is allowed to manage.

Teng framed Virtuals as a decentralized alternative to agent ecosystems being explored by traditional financial institutions. He said agents that operate fully on-chain may avoid identity requirements until they connect with fiat payment systems or banking infrastructure, where traditional KYC rules are likely to apply.

The company is also looking beyond purely digital agents. CoinDesk reported that Virtuals is working with robotics startups, universities and Balaji Srinivasan’s Network School ecosystem to test real-world humanoid applications, with Virtuals focused on software, commercialization and data collection rather than building robot hardware.