Orkes Raises $60M to Orchestrate Enterprise AI Agents in Production

When Jeu George talks about AI in the enterprise, he does not start with model architectures or benchmarks. He talks about outages. In the past year, the Orkes co-founder says, too many companies have learned the hard way that a clever large language model demo does not translate into a reliable production system that can survive rate limits, flaky APIs, or a compliance audit. Orkes, the workflow and agent orchestration startup he leads, is betting that this reliability gap is where the real money in AI will be made—and its new 60 million dollar funding round suggests investors agree.

Inside Orkes’ $60 Million Bet

Orkes, based in Santa Clara and built by the original architects of Netflix’s microservices orchestration platform, has secured 60 million dollars in fresh capital led by AVP, with participation from Prosperity7 Ventures, Nexus Venture Partners, Battery Ventures, and Vertex Ventures US. The round, which includes 40 million dollars in Series B equity, brings Orkes’ total funding to roughly 90 million dollars and places it among the more heavily financed players going after enterprise AI orchestration.

Investors are effectively underwriting the idea that enterprises will not scale AI without a dedicated control plane. The company’s pitch is that Orkes gives developers a single execution layer to design, run, and monitor complex, AI‑infused workflows—everything from multi-step customer support agents to fraud pipelines that blend LLM reasoning, internal microservices, and human approvals.

From Netflix Conductor to Enterprise AI Backbone

Under the hood, Orkes builds on Conductor, the open‑source workflow engine first created at Netflix to manage sprawling microservices architectures. That lineage matters to risk‑averse CIOs: Conductor has already been battle‑tested at web scale, orchestrating millions of transactions across services that cannot afford downtime.

Orkes has repurposed that foundation for an AI-first world. The platform lets teams stitch together LLM calls, retrieval steps, deterministic services, and human tasks into durable workflows, with retry logic, timeouts, circuit breakers, versioning, and observability built in. If a model times out or an external API fails, the workflow can degrade gracefully rather than dropping a customer request or producing an untraceable hallucination.

Why Developers Are Reaching for Orchestration

For developers, the problem is not building a prototype chatbot; it is running a fleet of AI agents that interact with core systems without breaking SLAs or compliance rules. Orkes pitches itself as the execution layer that abstracts away the mess: it tracks state across long‑running processes, provides detailed logs and metrics, and exposes policy controls so security teams can see which agents can call which systems and under what conditions.

In practice, that means a fintech can design a credit underwriting agent that chains together a KYC service, an LLM for document parsing, a rules engine, and a human reviewer, with every step logged and auditable. A retailer could run a customer support agent that escalates to a human when confidence scores drop below a threshold, with dashboards showing latency, error rates, and model costs over time.

This is the layer many organizations realized they were missing in 2024 and 2025, when AI pilots worked in a sandbox but stalled before hitting real users. Orkes is explicitly going after that “stuck in pilot” segment: companies with budget and executive pressure to deploy AI, but without the tooling to make those systems robust enough for production.