You'll design and ship production multi-agent systems for enterprise clients, the AUTOMATE layer of our Agentic Growth Stack. That means orchestrating fleets of LLM agents that qualify leads, draft outreach, coach revenue teams, and write back to CRM, on a reliability bar where they actually run in production for months without losing the plot.
You'll own the orchestration topology, the tool ecosystem, the agent memory architecture, the human-in-the-loop checkpoints, and the observability stack. As the senior anchor for agentic engineering depth in Lisbon, you'll also be the buddy and senior reviewer for our Tbilisi Agentic Engineer when that hire ramps later in the year.
This is engineering, not research. We ship.
You'll ship 3+ production multi-agent systems to enterprise clients within your first 12 months (deployed, signed-off, in active use as AUTOMATE-layer engagements)
You'll build and maintain our reusable agent orchestration framework: orchestrator pattern, tool-calling layer, agent memory layer, human-in-the-loop hooks, observability hooks, usable by every future Agentic Engineer across all sites
You'll establish our agent observability and reliability stack: tracing, cost dashboards, drift monitoring, error-recovery and fallback patterns, with documented SLAs per agent type
You'll establish our agent quality bar: every shipped multi-agent system has documented orchestration-level evals (not just per-tool unit evals), a rollback plan, and an incident playbook
You'll pair daily with the Tbilisi Agentic Engineer once that hire ramps, with the explicit goal that Tbilisi takes ownership of at least one production agent within 6 months of joining
You'll co-author 2+ public-facing technical assets (blog post, webinar, conference talk) on our multi-agent architecture approach
You'll contribute as the agent architect in pre-sales scoping alongside Sales, Advisory, and the Head of AI when in seat
Must-haves
Production experience shipping multi-agent systems beyond prototypes (orchestrator-plus-sub-agents, supervisor patterns, agent-to-agent handoffs). You can walk us through what you shipped, what broke, and how you fixed it
Hands-on production experience with at least one major agent framework (LangGraph, LangChain, AutoGen, CrewAI, Semantic Kernel, Pydantic AI, Mastra, or comparable), and a credible point of view on why one over another for a given engagement
Track record of long-running agent loops (10+ steps) holding up in production: state management, retry policies, max-step caps, looping detection, graceful recovery from tool failure
Production fluency with agent observability and tracing (LangSmith, Helicone, Arize, Langfuse, or self-built). You read traces like an SRE reads CPU
Tool design instinct: you've tuned tool schemas and docstrings based on observed model behaviour, and you know when a tool should be one tool vs. several vs. a sub-agent
Strong Python or TypeScript with tests, types, CI, deployment in cloud (AWS, GCP, or Azure), Docker, and basic IaC
Native-level Portuguese plus business-fluent English
Nice-to-haves
Public open-source contributions to agent frameworks (LangGraph, LangChain, CrewAI, etc.)
Shipped commercial multi-agent products at scale (B2B SaaS, agentic platforms, enterprise AI products)
Direct HubSpot or comparable CRM integration experience
Eval engineering background at the orchestration level (not just unit-level model evals)
You'll report directly to the Director Development & Technology, with a dotted line to the Head of AI & Agentic Delivery once that role is in seat. You'll join the AI Team as the Lisbon anchor for agent orchestration. Day-to-day, you'll work closely with the AI & Data Engineer (single-agent and data foundation), the Director D&T (technical sign-off and architecture review), and the broader Lisbon hub. Once the Tbilisi Agentic Engineer joins later in the year, you'll be their buddy and senior reviewer, with a real ownership transfer goal at the 6-month mark.
Languages: Python and TypeScript
Agent frameworks: LangGraph, LangChain, AutoGen, CrewAI, Pydantic AI, Mastra (we pick per engagement)
LLMs: Anthropic Claude (we are an Anthropic Build Partner), OpenAI, Gemini, with self-hosted options where the engagement demands it
Observability: LangSmith, Helicone, Langfuse, Arize
Cloud and DevOps: AWS, GCP, Azure, Docker, GitHub Actions, basic IaC
CRM and integration: HubSpot APIs (REST, GraphQL, webhooks)
Collaboration: Jira, Confluence, Forecast, Claude AI, Claude Code, Cursor
Statutory: Standard Portuguese employment benefits via local entity or EOR (paid time off, public holidays, parental leave, statutory health coverage)
Health: Private health insurance top-up
Learning: Full Blinkist Business library (4,500+ books), 3 months of Babbel, dedicated AI conference and training budget
Flexibility: Up to 4 weeks per year working from anywhere in the EU with a €500 allowance, hybrid setup with Lisbon hub access
Culture & Tools: Flat hierarchies with direct access to CEO and Leadership, modern stack (HubSpot, Jira, Confluence, Claude AI), Anthropic Build Partner status with early access to Claude capabilities
Thorit is one of Europe's leading HubSpot partners and a top business and technology consultancy. We combine strategic advisory with hands-on IT implementation, from CRM architecture to fully automated go-to-market infrastructure powered by AI and agentic systems.
Send your application via Ashby. A cover letter is optional, but a concrete example of a multi-agent system you have personally shipped to production (architecture, framework choice, what broke and how you fixed it) is required. Public links (GitHub, blog post, demo video) appreciated.
