Agentic AI Engineer
Build the agent infrastructure that lets the rest of the studio ship faster. Design multi-agent pipelines, write Claude skills, and solve the production problems that most agent demos never reach.
The gap between a working agent demo and an agent system that ships to production and stays there is almost entirely about the things that never appear in the demo. Failure modes the happy path never touches. Evals that catch a regression before a user does. Context windows that fill up with the wrong things. Guardrails that are not theater. A human-in-the-loop flow that does not just add latency while pretending to add oversight. This role is for the person who has hit those walls, built around them, and ended up with something that actually ran.
We build agentic systems for clients and into our own engineering workflow. The internal approach is documented in our agentic coding guide and in our thinking on why most agentic AI stalls before it reaches production. You will be working in both directions: building multi-agent pipelines, writing Claude skills and MCP integrations that the rest of the engineering team uses day to day, and solving the production problems that appear when a system designed in a notebook meets real load, real users, and real constraints. You will also define what the guardrail and eval standards look like for the studio as a whole, because that playbook is genuinely not finished yet.
Idealogic's 2028 strategy names shipping AI to production rather than to pilots as a central goal. This role is how that gets done at the infrastructure level: the pipelines, the tooling, the standards that let the rest of the team ship agentic work without reinventing the safety checks from scratch on every project.
The work is broad and the problems are genuinely unsolved for most of the industry. If you want to build the agent systems rather than just use them, and you have the production experience to show you have done it before, send your CV and tell us about one that shipped.
Responsibilities
- Design and build multi-agent pipelines, orchestration layers, and the infrastructure that makes agentic systems reliable in production rather than just impressive in a demo
- Write Claude skills, tools, and MCP integrations that let the rest of the engineering team work faster and with less friction
- Take an agentic system from a proof-of-concept that works on the right inputs to something that holds up under real load and real failure modes
- Define the guardrails, eval frameworks, and review gates that separate a safe agent from one that writes its own requirements
- Work across client projects and internal tooling, so the lessons from production flow back into the tools the studio uses every day
Requirements
- Five or more years in a senior engineering role, with direct experience shipping systems to production rather than to staging
- You have built with large language model APIs in earnest, not just called a chat endpoint and wrapped it in a React component
- Deep understanding of agent architectures, including tool use, multi-agent coordination, context management, and the failure modes that documentation does not cover
- You have hit the ceiling of what an LLM does reliably and you know how to design around it, which means evals, guardrails, human-in-the-loop flows, and honest scope
- Comfortable with TypeScript and the surrounding Node.js ecosystem; the agent work is the hard part, not the glue code around it
Nice to have
- Shipped a multi-agent system in a regulated domain or in a context where a failure had real consequences
- Written Claude Code skills, MCP servers, or custom tooling for a coding agent
- Contributed to open projects in the agent tooling or AI infrastructure space
What we offer
- Work at the leading edge of how software gets built, not as a future thing but as the actual present
- Ship agent systems to real clients and into the studio's own delivery workflow, so the work compounds rather than staying on a shelf
- Remote-first across Europe, senior teammates, and a culture where the interesting problems are not parceled out by seniority
- A direct line to the strategic direction in Idealogic's 2028 plan, where agentic AI in production is a named goal
- Room to define how the studio builds with agents, because that playbook is still being written
Think this is you?
Send a short note and your CV. We read every application — no keyword filter, no black box.
Other open roles
More ways to build with the team across engineering, design, and product.