Software Engineer/Research Engineer, Agent Orchestration

About the Team

The Agent Orchestration team builds the runtime and model orchestration layer that powers Decagon’s agents in production. This is the orchestration layer that turns workflows, tools and guardrails into a reliable, low-latency, and delightful experience for end users.

At the core of this work is the agent harness: the routing, execution logic, tool orchestration, and control-plane systems that determine how an agent behaves in a live conversation. The team owns the full execution lifecycle of each conversation—from selecting workflows and orchestrating multiple models (e.g., router/planner/supervisor patterns), to coordinating tool calls, enforcing safety constraints, and communicating back to the user.

The team operates across both real-time systems (e.g., voice interactions with strict latency requirements) and longer-horizon execution (supporting more complex reasoning and workflows). Our research shows that an agent’s task execution reliability increasingly depends on the orchestration layer that wraps around it.

This is highly experimental, frontier-style engineering. The team continuously analyzes real-world failures, builds feedback loops through offline evaluation and online experimentation, and iterates quickly to improve quality, reliability, and capability. As model capabilities evolve, the team regularly rethinks system design to push agent performance forward in production.

About the Role

As a Senior Software Engineer on the Agent Orchestration team, you will design and build the systems that govern how Decagon agents operate in real-world environments.

You will own complex, distributed systems that sit at the heart of the agent runtime: execution frameworks, model orchestration logic, and experimentation platforms that ensure agents are fast, reliable, and continuously improving. Your work will directly impact how agents reason, take actions, and deliver outcomes across millions of interactions.

This role operates in a fast-moving, ambiguous space with tight feedback loops. You’ll move fluidly between diagnosing production issues, designing new system abstractions, and running experiments to improve agent behavior. You’ll collaborate closely with Research, Infra, and Product teams to ship improvements safely and at scale.

In this role, you will

Design and evolve agent harnesses that power different product experiences

Build core runtime systems, including AOP execution and multi-model orchestration

Develop control-plane logic for routing, planning, and tool invocation with strong safety guarantees

Optimize agent systems for latency, reliability, and production correctness

Analyze real-world failures and use data to drive iterative improvements

Build and operate online experimentation (A/B testing) and contribute to offline evaluation frameworks

Improve observability, testing, and simulation systems to ensure safe, measurable progress

Contribute to voice and real-time systems (e.g., transcription pipelines, turn-taking, latency improvements)

Continuously adapt orchestration systems as model capabilities evolve

Your background looks something like this

Strong experience building distributed systems or backend platforms in production environments

Comfort working in ambiguous, fast-moving environments with rapid iteration cycles

Experience owning systems end-to-end, from design through production and iteration

Familiarity with experimentation, evaluation, or data-driven product improvement loops

A track record of improving system reliability, performance, and observability

Ability to debug complex systems and identify root causes of failures

Even better

You’ve built or worked on agent harnesses, orchestration layers, or execution frameworks

You think in terms of control planes, feedback loops, and system-level optimization, not just features

You’re excited about diagnosing failure modes and iterating toward measurable improvements

You care deeply about production quality—not just making systems work, but making them reliable, safe, and scalable

You’re motivated by pushing the frontier of how intelligent systems behave in the real world