Breaking Down Somnia’s April Pivot

Somnia's April pivot was a product thesis shift: from high-throughput infrastructure to an execution environment for agents.

How the shift took shape

On March 23, Somnia brought in Peter Lipka as CEO, Harry Lang as CMO, Kevin Zia as COO, while Paul Thomas moved to long-term technical direction. Less than a month later, the Agentic L1 thesis went public. On April 29, Prophecy Social launched, the first application built on the assumption that an AI agent, not a person, is the primary actor.

Lipka co-founded Improbable and built SpatialOS, a system designed for massive concurrency and shared state at scale. That background maps pretty directly onto what Somnia is now trying to build.

Automated execution already shapes crypto markets. Somnia pushes more of that execution into the chain itself, where data access, decision logic, and settlement share the same environment.

The infrastructure was already there

Somnia was built for high-frequency interaction before the agent narrative existed. MultiStream consensus separates data production from global ordering, so validators publish their own data chains asynchronously while a separate consensus chain sequences the heads. IceDB runs at 15-100 nanosecond read/write with performance reports that keep gas pricing predictable under load. Accelerated sequential execution compiles EVM bytecode to native x86, rather than relying on parallelism that breaks down when many transactions hit the same state.

Agents create correlated load. They converge on the same opportunities, data, and contracts at the same moment. That is a more useful stress test than any clean throughput benchmark.

Somnia claims over 1,000,000 TPS with sub-second finality. The testnet logged more than two billion transactions, 10 million unique wallets, 50 validators, 70-plus ecosystem partners.

What changes onchain

Smart contracts on most chains are passive. Something has to call them. Automated behavior means keeper bots, polling systems, external webhooks, all off-chain, all introducing trust assumptions.

The April 21 announcement said Somnia Agents are running onchain "as part of consensus." In practice that means:

• Smart contracts can natively query external APIs
• Agents can run deterministic AI inference inside the execution environment
• Web data can be scraped with results passing through consensus validation
• Reactive smart contracts respond to emitted events without external triggers

Reactivity is still testnet-only as of May 2026. A chain for agents has to notice when execution is needed rather than wait to be called, and that piece isn't fully live yet.

Prophecy Social

On April 29, Somnia launched Prophecy Social, a prediction market platform where AI agents resolve outcomes.

Polymarket relies on UMA's optimistic oracle model, where outcomes can be proposed and disputed. That approach is robust, but it is not designed for second-by-second resolution. Prophecy Social uses agents that analyze API data and resolve markets almost immediately, which makes faster-resolving, more granular event markets viable: short-lived outcomes, live-event states, creator-driven markets, cases where slow oracle resolution just breaks the experience.

Paul Thomas called it a "market of markets." Autonomous loops with settlement is a precise way to describe the kind of workload Somnia is targeting.

What makes the pivot feel credible is that Somnia already has a real first application for it. Prophecy Social puts agents exactly where the thesis gets tested: speed matters, external data feeds in, resolution happens onchain. From our side, Prophecy Social is the part worth watching closest. It puts Somnia's agent thesis into a live workload, external data in, AI resolution, onchain settlement, and if that loop holds under real conditions, the architecture claim starts to mean something.