AI Systems / Agent Engineer- SOUTH AFRICA (REMOTE)

Mission: Build the engineering harness that allows a small specialist model to operate reliably as an end-to-end software engineer.

Responsibilities

• Design and build the agent/harness architecture around specialist models.
• Build repository understanding and codebase retrieval.
• Connect models to Git, compilers, debuggers, test runners and development environments.
• Build planning → execution → testing → evaluation → correction loops.
• Integrate architecture/dependency analysis.
• Integrate security and static-analysis tooling.
• Build automated performance/load-testing workflows.
• Design model verification and critic/evaluator systems.
• Develop tool-use and function-calling infrastructure.
• Build mechanisms that prevent models from modifying code before understanding the task/context.
• Develop observability and evaluation infrastructure for agent behavior.
• Determine which capabilities should be handled by the model and which should be deterministic system components.
• Work closely with the Model Research Engineer to feed real-world failures back into model training.

Ideal background

• Strong AI agent / ML systems engineering experience.
• Experience building production coding agents or developer tools.
• Strong Python/TypeScript and backend engineering.
• Experience with tool calling, orchestration and agent frameworks.
• Strong understanding of software architecture.
• Experience with RAG, code retrieval and knowledge graphs.
• Familiarity with Docker, Kubernetes, CI/CD and cloud infrastructure.
• Experience with evaluation and observability systems.

Bonus

• Compiler/toolchain experience.
• Static analysis.
• Code security.
• Performance engineering.
• SWE-bench or repository-level coding agents.
• LangGraph, Claude Agent SDK, OpenAI Agents SDK or similar.

What success looks like

Take a small specialist model and build a system around it that allows it to reliably understand, plan, implement, test, debug and validate real software-engineering tasks.