Restate Builds Durable Infrastructure for AI Agents

Restate is building the durable runtime underneath AI agents, workflows, and distributed backends. The Berlin-based developer-infrastructure company was founded in 2022 and is led by co-founder and CEO Stephan Ewen alongside co-founders Till Rohrmann, Igal Shilman, and Ahmed Farghal. Their goal is direct: make durability, state, communication, and coordination feel like ordinary programming primitives.

That mission matters because modern software does not fail in clean, convenient places. An agent can make hundreds of model calls, use external tools, wait for human approval, and change direction while work is still running. If the process crashes halfway through, restarting from the beginning can repeat side effects, lose context, and turn a recoverable error into an operational incident. Restate records progress so execution can resume from the unfinished step.

Restate Treats Durability as a Runtime Primitive

Restate's runtime sits between application code and the failure mechanics that usually spread across queues, databases, retry loops, locks, and state machines. Developers keep writing services in familiar languages while Restate journals completed work, coordinates calls, stores keyed state, and recovers execution after a failure.

The product is broader than a workflow engine. Restate combines Durable Functions, Workflows, Virtual Objects, reliable RPC, queues, timers, signals, and flow control. Those components can compose across AI agents, transaction systems, control planes, and event-driven backends rather than forcing every workload into one workflow-and-activity model.

The architecture is the sharper part of the thesis. Restate uses a distributed log, RocksDB for local storage, and object storage for durable snapshots. The company says this design keeps the hot path simple and reduces coordination overhead. Restate advertises less than 10 milliseconds of durable-action overhead under load, more than 100,000 actions per second in production, and costs up to 10 times lower through its managed BYOC offering. Those are company-reported performance claims, but they show the boundary Restate is trying to move: durability from a special-purpose safety layer into the normal path of application execution.

AI Agents Made the Failure Problem Impossible to Ignore

Durable execution existed before the agent boom. The timing changed because agents made long-running, asynchronous programs common. Model calls time out. Tool APIs fail. Human approvals arrive hours later. A process may survive several deployments before it completes.

Checkpointing one agent loop is not enough when the surrounding services, messages, state, and tool calls can still drift apart. Restate's market argument is that the whole interaction needs a durable execution model. That includes the work already completed, the signals still pending, and the side effects that must not run twice.

This is also why Restate's $20M Series A is more than a financing headline. Singular led the September 2026 round, with Redpoint Ventures and Capital One Ventures participating. The round brings disclosed funding to $27M and gives Restate more room to hire engineers, build a go-to-market organization, and expand its San Francisco presence as infrastructure buyers move agents from demos into production.

Production Use Is the Real Test

Restate points to a deliberately varied set of production users. Replit uses Restate as durable orchestration for its coding-agent platform. DOSS replaced an internal BullMQ and Postgres workflow runtime with Restate. A Fortune 500 U.S. bank uses the platform for treasury workflows that require multi-region consistency. Bilt uses it for transaction processing.

The workloads differ, but the operating problem is the same. Teams need software to recover without losing completed work, duplicating a payment, or leaving a user session in a half-finished state. Restate says hundreds of companies now use the product, while TechCrunch reported multiple six- and seven-figure customer contracts based on company disclosures. The revenue figures are not audited, but the mix of agent, enterprise-resource-planning, and financial workloads is useful evidence that durable execution is escaping its original workflow niche.

Competition makes the category more credible, not less. Temporal announced a $550M Series E at a $12.55B valuation in September 2026 and said it serves more than 4,300 customers. Restate is much smaller, but its differentiation is clear: a self-contained runtime, composable service models, low-latency ambitions, and deployment through self-hosted software, Restate Cloud, or managed BYOC.

The Founding Team Has Seen This Movie Before

Ewen, Rohrmann, and Shilman worked on Apache Flink and at Data Artisans, later acquired by Alibaba and renamed Ververica. Their earlier work helped make stateful stream processing a reusable building block. Restate applies a related instinct to general backends: stop rebuilding recovery, consistency, and coordination inside every application.

The current team is compact and specialized. Farghal brings large-scale infrastructure experience from Meta and Stripe. Field CTO Jamie Grier adds a technical customer-facing role, while Head of Business Operations Jan Rixgens and VP of Sales Shane Nordstrand reflect the commercial organization now forming around the product.

Restate's careers page lists roles across engineering, design, developer relations, product marketing, and solutions architecture in Europe and the United States. The company emphasizes ownership, small teams, developer-first decisions, and quality because customers place Restate directly in critical execution paths. Hiring here is not simply a growth slogan. It shows the company building both the product depth and field capacity required for infrastructure adoption.

Why Restate Matters Now

The agent market has spent plenty of time celebrating what software can attempt. Production teams are now paying for what happens when that attempt breaks at step 73 of 100. Reliability is becoming part of the agent runtime, not a cleanup project assigned after launch.

Restate's opportunity is to make durable execution boring in the best possible way: present everywhere, noticed mainly when it prevents failure, and simple enough that teams stop assembling their own recovery machinery. The company still has to prove its performance claims across a much larger customer base and compete with a heavily funded incumbent. But its product design, founding experience, production mix, and current hiring point toward the same conclusion. As agents become longer-running and backends become more stateful, recovery is moving from an implementation detail to core infrastructure.