Orkes runs the code you're afraid to ship.

Selling the part of AI nobody demos

There is a genre of technology company whose product you can only appreciate by its absence. You do not admire orchestration the way you admire a chat interface or a slick dashboard. You notice orchestration at 3am, when a workflow that was supposed to move money from one place to another silently stopped halfway, and nobody can say where. Orkes, a company based in Cupertino, California, is in the business of making sure that phone call never happens. This is a less glamorous pitch than "we have built an intelligence," but it may be a more durable one.

The setup is worth appreciating, because it is the kind of origin story that startup investors like to underline twice. Back in 2016, a group of engineers at Netflix had a problem that is very Netflix: thousands of microservices, each doing its own small job, and no reliable way to make them cooperate on a larger one. So they built a tool called Conductor - a workflow engine that could stitch these services together, keep track of where a long-running process was, and pick it back up if something failed midway. Conductor worked. It became open source. It got adopted well beyond Netflix, by companies that had the same problem and no desire to solve it from scratch.

And then, in 2021, several of the people who built it - Jeu George, Viren Baraiya, Boney Sekh, and Dilip Lukose - did the thing that is either obvious or brave depending on where you sit. They left, and founded Orkes to turn the free engine into a business. George is chief executive; Baraiya, who created much of the original Conductor, is chief technology officer; Sekh is co-CTO; Lukose is chief product officer. The premise is not that they would invent something new. The premise is that they had already invented the thing, thousands of companies were using it for free, and some meaningful fraction of those companies would happily pay someone else to run it for them.

The unglamorous word: durability

If you want to understand Orkes in one word, the word is "durability," and it means something specific. A durable workflow is one that remembers where it was even when the thing running it disappears. The server crashes; the process resumes. A step fails; it retries, or it waits, or it hands off to a human for approval and then keeps going. This sounds like a small feature until you are the person whose payment pipeline died in the middle of a transaction, at which point it is the only feature that matters. Orkes sells that property, packaged, hosted, and wrapped in the things enterprises insist on: role-based access control, audit logs, observability dashboards, and support contracts with someone to call.

The commercial mechanism here is what people in the industry call "open core," and Orkes runs a fairly clean version of it. The engine - now maintained as Conductor OSS, with Orkes as the primary steward - is free. Hundreds of thousands of developers use it, across millions of installs. That community is the top of the funnel. The money is in Orkes Cloud, the managed enterprise version you can deploy on the cloud of your choice, sold on subscription to the companies that cannot tolerate downtime and would rather not staff a team to babysit orchestration infrastructure. Give away the engine, sell the reliability. It is not a new playbook, but it is a proven one.

The rounds tell a consistent story. Orkes came out of stealth in 2022 with a $9.3M seed backed by Nexus Venture Partners and Battery Ventures. It raised a $20M Series A in early 2024 as the platform gained traction. And in April 2026 it closed a $60M Series B led by AVP, with new investor Prosperity7 Ventures joining Nexus, Battery, and Vertex Ventures US. Between the Series A and the Series B, the company says it tripled its customer base. The through-line: not a company selling a story about what orchestration could be, but one raising against a growing invoice list of enterprises already paying for it.

Why any of this suddenly matters more

Here is the part that turns a solid infrastructure business into a timely one. Everyone wants to deploy AI agents. Very few people want to talk about what happens when an agent, left to run a multi-step task involving real money or real patient data, fails somewhere in the middle and cannot tell you where. A brilliant agent that finishes the job 95% of the time is not a product. It is a liability with good marketing. The gap between an impressive demo and something a bank would actually put in production is enormous, and it is almost entirely a reliability gap.

Orkes's bet is that this gap is exactly the shape of the thing it already built. An orchestration engine that can run a long, failure-prone, multi-step process durably, retry the parts that break, and pause for a human when the stakes demand it, does not especially care whether the steps are microservice calls or LLM decisions. So the company has extended Conductor toward what it calls agentic workflows - structured process stages blended with model-driven decisions, tool use, and human-in-the-loop checkpoints. It has added Agentspan, an open-source durable runtime for AI agents, and an MCP Gateway that turns a company's internal APIs into safe, consistent tools an agent or an LLM is allowed to use.

The product, in four boxes

Who is actually paying

The customer list is the argument. The open-source Conductor project has been adopted in production by a roster that reads like a Fortune 100 index - and the enterprises buying Orkes Cloud span finance, healthcare, media, and telecom, precisely the industries where "the workflow silently failed" is a regulatory event, not an inconvenience. When the same orchestration engine underpins a streaming service, a car company, and an investment bank, the pitch to the next skeptical CTO gets shorter.

None of this makes Orkes a sure thing. The durable-execution category is competitive - Temporal, AWS Step Functions, Apache Airflow, Camunda, Prefect, and a wave of newer AI-agent frameworks are all reaching for adjacent ground. Open core is a real model but a demanding one: you have to keep the free thing genuinely good while convincing a fraction of its users to pay. And the reported revenue figures floating around are modest relative to nearly $99M raised, which means the company is being funded on trajectory, not on current receipts. But the founders felt this exact pain, at Netflix, at scale, before it was a market. The best infrastructure companies usually are scar tissue turned into software. Orkes is a fairly literal example.

From an internal tool to a company

Frequently asked