Principal Artificial Intelligence Engineer

Principal AI Engineer, AI-Native Software Development Freelance Contract
- one year plus extensions Location Brussels-50% onsite Role Summary A principal-level software engineer who has moved their own practice to agentic development and can bring an organisation with them, properly. You will define how software is built with AI agents across the organisation: repository conventions, agent context files, reusable skills and workflows, quality gates and integrations with the SDLC toolchain. You will also build the internal AI agents and applications the CoE delivers. Software engineering depth is the core of this role; AI is how you apply it.

Key Responsibilities
Engineering standards for AI agents
• Define and own the AI-native engineering standard: repository structure, coding conventions, and a layered agent-context model (CLAUDE.md / AGENTS.md at organisation, repository, module and path level) that produces consistent output across teams and tools.
• Build and maintain a shared, versioned library of agent assets (skills, custom commands, subagents, hooks, plugins, prompt templates) distributed to teams through a controlled catalogue.
• Enforce the standard deterministically, not only by instruction: hooks, pre-commit checks, CI validation of context files, golden-repository templates and scaffolding.
• Establish a spec-driven development practice (specification → plan → implementation → verification) aligned with the CoE's gated AI-first SDLC, with traceability from requirement to code, test and release. SDLC integration
• Design and build integrations between AI agents and the delivery toolchain (Jira, Confluence, ServiceNow, Git platforms, CI/CD), including MCP servers and API-based connectors.
• Automate high-value flows: backlog refinement and acceptance-criteria generation, spec and ADR drafting in Confluence, AI-assisted pull-request review, test generation, release notes, incident and change-record support in ServiceNow.
• Run agents headless in CI pipelines where appropriate, with scoped permissions and a full audit trail.
• Design, build and operate multi-step agent pipelines that automate SDLC stages end to end (e.g., ticket → spec → implementation → pull request → review), with human approval gates, persistent state, tracing and cost controls. Quality, safety and measurement
• Define the review and acceptance policy for AI-generated code, including where human review is mandatory.
• Implement quality and security gates suited to AI-generated code: SAST, dependency and licence scanning, detection of non-existent or typo-squatted packages, secrets scanning, coverage and mutation-testing thresholds.
• Build evaluation harnesses for agent workflows so that changes to models, skills or context files are regression-tested before rollout.
• Measure impact with engineering metrics (DORA, lead time, rework rate, defect escape rate, review load) and report honestly on where AI helps and where it does not. Build and enable
• Design and develop, hands-on, full-stack AI applications and agents for the CoE (backend, frontend, agent orchestration, retrieval, tool integration).
• Apply AI agents to modernisation of existing Java estates: codebase comprehension, framework and runtime upgrades, test backfill.
• Coach delivery teams, run workshops, review code and agent configurations, and grow a network of AI champions.
• Evaluate new models and coding tools in a vendor-neutral way and recommend adoption or retirement. Essential Requirements
• 10+ years of professional software development, including 3+ years at principal, staff or lead-architect level defining engineering standards adopted across multiple teams.
• Expert in at least one backend stack (Java / Spring Boot preferred, or Python) and proficient in TypeScript with a modern frontend framework (Angular or React).
• 2+ years building LLM-based applications that reached production: tool use / function calling, structured outputs, retrieval-augmented generation, context management and evaluation.
• Proven track record designing and building AI agents and multi-step agent pipelines that run in production: tool-calling loops, multi-agent orchestration (orchestrator–worker, handoffs, parallel subagents), human-in-the-loop checkpoints, state and memory management, error handling and retries, guardrails, and tracing of agent runs. Hands-on with at least one agent framework or SDK (Claude Agent SDK, LangGraph, Semantic Kernel, OpenAI Agents SDK or similar).
• At least 12 months of daily, hands-on use of agentic coding tools on real codebases, including Claude Code. Demonstrable command of context files, skills, subagents, hooks, MCP servers and headless/CI usage; working knowledge of comparable tools (GitHub Copilot, Cursor, OpenAI Codex or similar).
• Proven integration work with Atlassian (Jira, Confluence) and ServiceNow APIs; experience building MCP servers or equivalent tool integrations.
• Strong engineering fundamentals: test-driven development, clean architecture, code-review culture, ADRs, trunk-based or equivalent branching, CI/CD.
• Secure SDLC knowledge, including OWASP Top 10 and OWASP Top 10 for LLM Applications (prompt injection, excessive agency, sensitive-information disclosure).
• Enough DevOps capability to containerise, build pipelines for and deploy what you build (Docker, Kubernetes, GitLab CI / GitHub Actions / Azure DevOps).
• Excellent written and spoken English (C1+); able to set standards with senior architects and win over sceptical developers. Desirable
• Red Hat OpenShift; deployment on Azure or AWS; Terraform basics.
• Long-running or autonomous agents: durable execution, long-term agent memory, and trajectory-level evaluation (LLM-as-judge, replay of agent runs).
• Experience in aviation / ATM or other safety-critical or regulated environments; awareness of software assurance practice (e.g., ED-153).
• Familiarity with the EU AI Act and ISO/IEC 42001.
• Public evidence of practice: open-source repositories, published agent configurations, talks or articles on AI-assisted engineering.
• French.