Platform Engineer - Agentic AI & Harness Engineering

About the role

Kyndryl is hiring a Platform Engineer to build its internal agent harness: the runtime, orchestration and guardrails that turn foundation models into AI agents doing reliable, governed work at scale. The harness is treated as a product for teams building and running agents, and it must be model-agnostic and versioned so it survives model upgrades.

Key responsibilities

• Design and build the agent harness: runtime, orchestration, tool and MCP integration, context and memory, evaluation, guardrails and observability.
• Define golden paths and reusable patterns so teams ship reliable, governed and measurable agents instead of one-off scripts.
• Define the oversight model: policy-as-code, human-in-the-loop checkpoints, loop and failure detection, evaluation harnesses and cost control.
• Integrate agents with CI/CD, infrastructure, cloud services, observability and ticketing via APIs and MCP.
• Make agentic productivity measurable by defining, instrumenting and evaluating each use case.
• Keep the harness model-agnostic and versioned, and mentor teams while reducing friction between Dev, Ops, Security and adopters.

Required profile

• 5+ years in Platform Engineering, DevOps or SRE, building developer platforms or runtime and orchestration systems in production.
• Hands-on experience building agent harnesses or agentic systems (runtime, tool and MCP integration, orchestration, evaluation and guardrails).
• Experience treating a platform as a product: contracts, SLOs, adoption metrics and real users.
• Professional working English and Spanish.

Required skills

• Solid software engineering in Python, with the depth to build production systems and abstractions.
• Cloud-native background: Kubernetes, containers, Infrastructure as Code, and one major cloud (AWS, Azure or GCP).
• Understanding of how LLM agents fail and how to make them reliable.
• Nice to have: LangGraph, CrewAI, AutoGen, the Model Context Protocol, AI-assisted coding tools, LLMOps, agent evaluation or agentic observability, and prior work on harnesses stable across model upgrades.

Success in the first year includes teams shipping agents through golden paths, agentic workflows running in production with guardrails and measured impact, and the harness absorbing at least one model or provider upgrade without a rewrite. Kyndryl offers a hybrid-friendly culture, career development tools, learning opportunities including certifications, and well-being programs.