Overview
Episode Title: Orkes and Agentic Workflow Orchestration with Viren Baraiya
Podcast: Software Engineering Daily
Date: October 2, 2025
Guests:
- Viren Baraiya: Founder and CTO of Orkes, creator of Netflix Conductor
- Host: Gregor Vand
Main Theme:
This episode explores the evolution and challenges of workflow orchestration in modern, microservices-heavy software architectures. Viren Baraiya discusses his journey from building Netflix Conductor to founding Orkes—an enterprise-scale platform for agentic workflow orchestration that extends Conductor's open source foundation with AI and compliance-focused features.
Viren Baraiya's Background and the Genesis of Conductor (02:01–04:25)
- Viren’s career path included stints at Goldman Sachs (investment banking tech), Netflix (infrastructure/platform engineering), and Google (developer products).
- At Netflix, his team faced major challenges in “coordinating work across different microservices,” which prompted the creation of Conductor.
- Traditional event-driven systems (like service buses) were brittle, especially during failures, making coordination and troubleshooting difficult.
- Quote:
“The major issue...is not when things are working fine, but when things are not working fine...you have to talk to 20 different people, manage hundreds of different queues.” — Viren (05:00)
- Quote:
- Conductor aimed to bring “order to the chaos without sacrificing the benefits” of microservices.
The Philosophy and Mechanics of Workflow Orchestration (07:51–10:35)
What is Workflow Orchestration?
- Orchestration coordinates work across systems—whether containers, data pipelines, or services.
- Workflow engines offload state management, making distributed systems resilient, scalable, and easier to develop.
- Quote:
“Everybody is building a state machine one way or other...The hardest part...is how to maintain the state in a way that it remains consistent and coherent with business goals.” — Viren (08:54)
- Quote:
Rule-Based vs. Programmatic Orchestration
- Rule-Based Engines: Historically allowed business users to define orchestration rules.
- Good for simple flows, but break down with complexity; developers often had to translate rules into code.
- Conductor Approach: Programmatic, DAG-based (Directed Acyclic Graph) workflows; logic is implemented in real code, matching how developers think.
- Quote:
“If you can write code, you should be able to define workflow. It should be one-to-one; there should be no missing cases there.” — Viren (12:12)
- Quote:
From Netflix to Orkes: The Evolution of Conductor and Open Source (13:21–15:31)
- Conductor started as an internal Netflix project, later open sourced for broader industry benefit.
- As external adoption grew, ex-Netflix engineers (including Viren) founded Orkes to provide enterprise support and continued innovation.
- The open source Conductor project was moved out from under Netflix’s official umbrella, giving the community and Orkes more control.
- Quote:
“Let’s take it out of Netflix umbrella and put it into its own project repository. Giving community more control over the project and increasing the velocity.” — Viren (14:00)
- Quote:
- Orkes operates an "open core" model: open source base, with enterprise features (security, compliance, governance, support).
Orkes Platform: Features and Evolution (15:31–19:50)
-
Focused on making Conductor “enterprise ready”—support for highly regulated industries, deployment flexibility (SaaS, BYOC, on-prem).
-
Extended Conductor beyond asynchronous workflow:
- Added support for long-running workflows (months/years).
- Enabled low-latency, synchronous microservice orchestration (finishing in milliseconds).
- Built comprehensive integrations with foundational AI models and platforms.
-
Orkes addresses operational needs often unmet by open source alone (monitoring, reliability, performance, compliance).
- Memorable Moment:
“Our first few customers were open source adopters. [They] said, ‘Glad you guys started the company—can you help us?’” — Viren (19:10)
- Memorable Moment:
Agentic Orchestration & AI Integration (19:50–26:32)
- Orkes evolved into an “LLM orchestration platform”—allowing deterministic and non-deterministic workflows by integrating Large Language Models (LLMs).
- Not just deterministic flows; workflows can include LLMs/agents, so even the same inputs can produce different paths.
- Built extensive support for all major foundational AI/LLM models.
What is an “Agent” in this Context? (24:07–26:32)
-
The term “agent” is broad but typically refers to an autonomous LLM or system that can plan and execute towards a goal.
- Could include single or multi-agent systems (multiple LLMs and/or humans-in-the-loop; each with distinct responsibilities).
- Example: AI-powered code orchestration tools, where agents generate, test, compile code and developers can approve/commit.
-
Emphasis on “non-determinism” and flexibility—agents make real-time decisions rather than following fixed scripts.
- Quote:
“Agents by definition have some level of autonomy and, therefore, non-determinism built into it.” — Viren (24:32) - Quote:
"If you break it down...you focus on writing tools...But instead of...putting them together, an LLM is taking your input and deciding on the fly, how should I do this." — Viren (28:12)
- Quote:
Trust, Guardrails, and Enterprise Confidence in Agentic Systems (29:29–37:11)
-
Orkes prioritizes “trust and safety,” essential for enterprise adoption.
- Guardrails: Points where humans (or other agents/automated systems) validate critical decisions (e.g., before destroying a Kubernetes cluster).
- Developers explicitly define when/where guardrails trigger—“adding guardrails is a deterministic step.”
- Full transparency: Step-by-step execution graphs, inputs/outputs, and rationale for decisions (“Why was a claim denied? Not just ‘the AI said so.’”).
- Strong audit trails and access control; only authorized agents/users can access or invoke critical tools.
-
These measures help businesses and non-technical stakeholders “feel more at ease.”
- Quote:
“With the right level of access control, visibility, and human guardrails, we think that's going to be enough for someone to say, hey, we can trust the system.” — Viren (36:40)
- Quote:
Usability and Views for Diverse Stakeholders (37:11–38:32)
- Orkes enables role-appropriate visibility:
- Developers see full step-by-step technical details.
- Ops/business users get high-level overviews, with the option for business-specific dashboards.
- Quote:
“As a developer you are able to see every step...as an ops person, either you can look at the high level block...You can build very business specific views.” — Viren (37:49)
Interoperability: Conductor, MCP, and Protocols (38:32–41:39)
- MCP (Multi-Modal Communication Protocol) simplifies how LLMs discover and interact with tools/APIs.
- Orkes supports bringing your own MCP servers; is expanding out-of-the-box integrations with popular enterprise services (Outlook, Twilio, etc.).
- Orkes is building its own Conductor MCP server to help generate entire workflow graphs from goals (advancing towards more autonomous systems).
- Future support planned for agent-to-agent (A2A) protocols and other evolving standards.
Getting Started and Developer Experience (41:39–43:50)
- Fast Onboarding:
- Developers can test Orkes via templates for common workflow types (API orchestration, chat agents, business processes).
- Playground available online: developer.orkescloud.com
- Try prebuilt templates or start from scratch with no setup required.
- Three major “10 minute impact” categories:
-
Orchestrating APIs with system tasks
-
Building basic agent/chatbot flows (including multi-agent conversational demos)
-
Adapting business processes (order management, claims, etc.) using templates
-
Quote:
“Nothing beats like, you know, one click, go to this URL and start working on it.” — Viren (43:28)
-
The Future: AI-Driven Business Automation (44:24–45:42)
- The industry is fast-moving towards business users building agentic workflows (not just developers), powered mainly by describing goals/flows rather than designing end-to-end.
- Orkes is focused on:
- Further developing agentic workflow support.
- Enhancing trust, safety, compliance features.
- Bridging developers and business teams—no more restrictive DSLs or “quirky” rule engines.
- Investing in tools for business users to harness AI/LLMs securely and productively.
- Quote:
“I think they [rule-based workflows] are coming back, and I would say this time with a vengeance...just describe what you want to do and I'll figure it out and do it for you.” — Viren (45:08)
Notable Quotes (All with Timestamps)
- “The major issue...is not when things are working fine, but when things are not working fine...you have to talk to 20 different people, manage hundreds of different queues.” — Viren (05:00)
- “Everybody is building a state machine one way or other...The hardest part...is how to maintain the state in a way that it remains consistent and coherent with business goals.” — Viren (08:54)
- “If you can write code, you should be able to define workflow. It should be one-to-one; there should be no missing cases there.” — Viren (12:12)
- “Let’s take it out of Netflix umbrella and put it into its own project repository. Giving community more control over the project and increasing the velocity.” — Viren (14:00)
- “Our first few customers were open source adopters. [They] said, ‘Glad you guys started the company—can you help us?’” — Viren (19:10)
- “Agents by definition have some level of autonomy and, therefore, non-determinism built into it.” — Viren (24:32)
- "If you break it down...you focus on writing tools...But instead of...putting them together, an LLM is taking your input and deciding on the fly, how should I do this." — Viren (28:12)
- “With the right level of access control, visibility, and human guardrails, we think that's going to be enough for someone to say, hey, we can trust the system.” — Viren (36:40)
- “Nothing beats like, you know, one click, go to this URL and start working on it.” — Viren (43:28)
- “I think they [rule-based workflows] are coming back, and I would say this time with a vengeance...just describe what you want to do and I'll figure it out and do it for you.” — Viren (45:08)
Summary Table of Key Segments
| Timestamp | Topic | Key Points | |--------------|-----------------------------------------------|-----------------------------------------------| | 02:01–04:25 | Viren's background & Conductor origins | Building platforms at Netflix, state headaches, need for orchestration | | 07:51–10:35 | Orchestration Types & Philosophy | Rule-based vs. programmatic, state machines, developer-centric approaches | | 13:21–15:31 | Open source & Orkes founding | Moving Conductor out of Netflix, open core business model | | 15:31–19:50 | Orkes enterprise evolution | Features for reliability, compliance, AI/LLM support | | 19:50–26:32 | Agentic orchestration | Determinism vs. autonomy, multi-agent workflows, business tooling focus | | 29:29–37:11 | Trust, guardrails, and enterprise safety | Deterministic controls in agentic systems, explainability, audit trails | | 38:32–41:39 | MCP protocols, integration & extensibility | Tool exposure, cross-protocol support, agent-to-agent comms | | 41:39–43:50 | Developer onboarding & quickstart | Templates, playground, API/agent/business use cases | | 44:24–45:42 | Future roadmap | Democratizing workflow automation, safe LLM/AI enablement for business users |
Conclusion
This episode offers a deep look at the evolution of workflow orchestration technology, centering on how Viren Baraiya’s work brought about tools for modern microservices at scale—from Netflix Conductor to the enterprise-grade Orkes platform. The discussion balances practical engineering challenges, product philosophy shifts (from rule-based to agentic, AI-driven orchestration), and concrete enterprise concerns like trust, compliance, and usability. It should be helpful for engineers, architects, and business leaders interested in the next generation of workflow automation and AI integration.
Quick Start: Try Orkes at developer.orkescloud.com (O-R-K-E-S) — one-click template or custom workflow builds.
