Somnia is now the Agentic L1, and Somnia Agents are running onchain, as part of consensus, right now. The next phase of the internet will be shaped by agents operating onchain. As AI continues to advance, we’re moving towards systems that don’t just execute logic, but can interpret, decide, and act in real time.
Agents operate continuously, transact in milliseconds, and coordinate across systems without pause. This opens up a new class of applications, from AI-resolved prediction markets to autonomous DeFi and onchain agent coordination.
Our New CEO Peter Lipka has laid out the keynote argument for why this shift matters and why the timing is right.
Inside the article:
- Why agents are a new kind of user that existing chains weren’t designed for
- How Somnia puts AI onchain as part of validator consensus, extending decentralisation and verifiability to AI itself
- What becomes possible when AI decisions in insurance, hedging, and prediction markets inherit the same guarantees as any onchain operation
- Why the infrastructure decisions being made today will shape the agentic economy for years
- What’s live on mainnet right now, and how to start building
Smart contracts gave us a deterministic internet, code that executes exactly as written, without needing anyone's permission. That's a genuine breakthrough. But there's a layer they were never designed to handle: judgment. Resolving a prediction market, interpreting a real-world event, pricing a novel risk, these still require a human, a committee, or a centralised oracle somewhere in the loop. It's a gap the industry has filled by quietly reintroducing the very middlemen crypto was built to remove - closed APIs, permissioned oracles and a handful of model providers who get to decide what "truth" looks like.
At the same time, the one technology that can exercise judgment at machine speed is advancing faster than any technology in human history. Generative AI is becoming more capable every month and builders everywhere are asking the same valid question -
How do we bring that capability onchain and preserve the guarantees crypto was built for?
That's the question that Somnia has now answered. Because the real story isn't that AI is powerful. It's that today, using AI inside a smart contract means importing a major trust assumption from outside the chain. We think there's a better path. And we’ve built it.
Agents are a new kind of user
Chains today were designed primarily for human users. Block times tuned to human attention. Fee markets that assume a person is deciding whether to transact. Governance cycles measured in weeks. That design made sense; humans were the users.
Agents operate differently. They transact in milliseconds, coordinate across systems continuously, and don't pause between actions. Supporting them well isn't a matter of incremental tuning. It calls for infrastructure designed with their workload in mind from the start.
Why Somnia is the Agentic L1
Somnia is the first L1 to put agents onchain as part of consensus itself. That's the shift. The principles that made blockchain valuable in the first place - decentralisation, transparency, immutability - now extend to AI. Agents don't sit alongside the chain, calling into it through bridges and oracles. They run inside it, with the same guarantees as any other onchain operation.
The engineering foundation was already there when I joined Somnia: up to a million TPS, sub-second finality, full EVM compatibility, and a runtime built for high-frequency autonomous workloads. Those capabilities exist because we’ve always pushed to put more onchain - and now we complete the picture by putting AI there too.
At Improbable, I spent years working on massively concurrent systems, and one lesson came up repeatedly: the layer underneath tends to define the ceiling of what's possible above it. Teams that try to patch around an infrastructure limit usually find they can't. The same principle applies here. Agents are the workload that makes purpose-built infrastructure matter again.
What's different about Somnia
A lot of great work is happening at the intersection of AI and crypto, much of it at the application layer, agents that hold wallets, services that let smart contracts call external models. These are useful building blocks, and we're glad to see the ecosystem maturing around them.
Our approach is complementary, and it's focused on a specific question: what if AI inference could happen inside consensus? On Somnia, agents can execute deterministically as part of validator logic. Smart contracts can incorporate AI inference and interact with external data with the same verifiability properties as any other onchain operation. The goal is to make AI a first-class citizen of the state machine, not an external dependency.
Anywhere an AI decision holds significant impact - be that financial or legal - this is the infrastructure it belongs on. Insurance products, hedging products, prediction markets - all can now have a trusted AI in the loop. The models run as part of validator consensus and the decisions are deterministic and verifiable.
The window is Now
The agentic economy isn't a distant future. Builders are making infrastructure decisions today, and those decisions tend to be durable. The chains that earn the trust of this first generation of agent builders will shape the category for some time to come.
I've always believed execution is what matters - show, don't tell. So here's what matters: this isn't theory. Somnia Agents are live on mainnet. They are running onchain, as part of consensus, right now. We are building on it ourselves, with more to come, and you can build on Somnia too.
Start here: somnia.network, and the documentation.
Working on something ambitious? Reach out to me directly.