Software Engineer, Agentic AI

What do we need

We are looking for an experienced and highly skilled Software Engineer (Agentic AI), to build the AI agent at the core of Dotmatics Luma platform. You will own the agent end to end: how it reasons, which tools it uses, what it remembers, how it recovers from failure, and how we know it is working. You will work in a small, high-trust engineering team alongside Product, scientists, and the wider Dotmatics organisation.

This is also an AI-native engineering role. Your success won’t be measured by how many lines of code you write. It will come from how well you decompose work for coding agents, give them the right context, tools, and constraints, verify their output with tests and review, and ship production software faster than a traditional senior engineer without lowering the quality bar. Coding agents are your default path for implementation, and you stay accountable for architecture, correctness, security, and review.

We want someone who thinks like the engineers who build frontier AI products: rigorous about quality, curious about how agents fail, and quick to adapt as models improve. You use AI tools daily to build AI tools, and you ship agents that work in production, not just in demos.

This position will require at least 1 day per week in office in Boston 

In this role you will get to

• Develop the agent itself: the loop, harness, and control flow connecting the model, tools, and state, including subagents, long-running tasks, checkpointing, and recovery.
• Practice context engineering: decide what goes into the context window, when to compact or summarize, and how to keep the agent coherent over long-running experiments.
• Design tools for the agent, with clear interfaces and descriptions, covering instrument orchestration, scheduling, data capture, and developer tooling such as GitHub, repositories, CI/CD, and workflows.
• Give the agent memory and state persistence so it learns from and builds on earlier work.
• Implement permissions and sandboxing that define what the agent can do on its own, what needs human approval, and how mistakes are contained.
• Defend the agent against prompt injection and untrusted content, with auditability and data isolation suited to regulated environments.
• Evaluate the agent end-to-end on real multi-step tasks, then debug its behavior by reading traces and finding root causes, fixing the harness, tools, or prompts to improve reliability, tool-use accuracy, cost, and latency.
• Adapt the harness as models improve, and remove scaffolding that is no longer needed.
• Run an AI-first development loop (scope, plan, delegate to coding agents, verify, ship), preferring agent execution over hand-writing whenever quality holds, and build and maintain the team's agent harness for development itself (project memory, skills, hooks, MCP, CI checks) so agents can safely do bounded engineering work.
• Measure and improve that workflow through throughput, defect rate, and review cost, and coach others on AI-native practices to raise the team's bar for agent-driven delivery
• Set up the architecture and infrastructure to run the agent reliably, and manage configuration, environments, and secrets
• Deliver the services and interfaces around the agent in Python, Node.js/TypeScript, and React, including streaming output and extension points such as MCP

We are looking for people who have around 10 years of experience in software engineering, a minimum of 3 years in AI, including hands-on experience building LLM-powered agents in production. Ideally you will have a degree or higher education in Computer Science, Engineering, or a related technical field, and an interest in life sciences, lab instrumentation, or automation.

The Key Skills we are looking for:

• AI-native engineering: autonomous coding agents (Cursor, Claude Code, Codex, or equivalent) as the primary way you build.
• Production agents: at least one LLM-powered agent shipped to real users, handling multi-step tool use, error recovery, and long-running tasks.
• Context engineering, memory, and retrieval: a long-running LLM application where you can explain what went into the window, when it was compacted, and how coherence and retrieval quality (RAG, vector indexes, reranking) were measured.
• Agent security: permissions, sandboxing, human approval gates, prompt injection defenses, and audit logging, tested against adversarial inputs.
• Tool and API design for models: tools or MCP servers built for model use, with tool-selection and call accuracy measured.
• Production engineering: Python with type checking, automated tests, and packaging in CI, plus Node.js/TypeScript and React services and interfaces, including streaming output, with monitoring, tracing, and managed configuration and secrets.

You may also have:

• Experience with agent frameworks such as LangGraph, or building custom harnesses from scratch.
• Working knowledge of Databricks and BI tools (Sigma, Power BI, Spotfire).
• Observability for LLM systems (tracing, cost and latency monitoring).
• Fine-tuning, prompt optimization, or model evaluation experience.
• Authored skills, hooks, or MCP servers that made agent-driven delivery repeatable for a team.

We are open to speaking to candidates with more or less experience than listed
The compensation for different levels or experience can be discussed with your recruiter