Role overview The Context & Harness Committee is moving from strategy to execution. We are seeking three hands-on AI Tech Leads to define and validate our context engineering standards and evaluation harness before company-wide rollout.
Core responsibilities
Lead the technical design and delivery of the assigned Context and Harness work streams
Review context formats, ontology models, harness proposals, and integration approaches.
Support pilot teams across MFBS divisions with context conversion, fixture authoring, and evaluation.
Define how GitHub, Notion, and Slack content maps into the agent context engine.
Establish guidelines for federated context routing and token-budget allocation across sources.
Contribute to an ontology proof of concept covering one domain and one workflow using YAML-based schemas.
Validate integration with internal or external agent platforms, including access and smoke testing.
Capture pilot evidence and contribute to the standard and company best-practices package.
Required skills
Context engineering skills
RAG and retrieval pipelines: Design advanced search and retrieval systems that supply models with accurate, relevant, and timely information.
Context-window and token management: Compress conversation history, prioritize useful context, and filter noise to optimize model attention and cost.
Memory architecture: Build short-term and persistent memory systems that allow agents to retain state across tasks and interactions.
Instruction curation: Structure plain-English operating rules, Markdown files, domain guides, and other instructions for dynamic injection into agent tasks.
Context formats and knowledge modeling: Design context layers and reusable knowledge assets using Markdown, YAML, schemas, and related formats.
Harness engineering skills
Tool and MCP integration: Connect models to external APIs, execution environments, tools, and Model Context Protocol (MCP) servers.
Sandboxing and permissions: Define safety boundaries, access controls, approval flows, and secure execution environments for agent actions.
Validation and guardrails: Implement linters, automated tests, evaluations, and verification layers that detect hallucinations, policy violations, or broken code.
Orchestration and control loops: Design multi-step workflows, retry logic, state transitions, and error-correction loops for long-horizon tasks.
Evaluation harnesses: Build fixtures, benchmark tasks, quality gates, and CI-integrated evaluations using DeepEval, Promptfoo, or comparable frameworks.
Supporting technical and leadership skills
LLM and agent fundamentals: Strong understanding of agent workflows, RAG, evaluation methods, and quality gates for AI-generated output.
Software engineering: Strong Git/GitHub and CI practices, experience with repo-native tooling, and the ability to prototype quickly.
Technical leadership: Ability to make pragmatic architecture decisions, lead POCs, and turn evidence into reusable standards.
Communication: Ability to write clear technical proposals and provide constructive cross-team review and feedback.
Experience with ontology or knowledge modeling using YAML-based schemas.
Familiarity with internal or external agent platforms.
Experience with hybrid search, indexing pipelines, or context freshness mechanisms.
Success outcomes
Context architecture, format, and evaluation standards are validated through representative pilots.
Pilot teams can convert source knowledge and author evaluation fixtures using documented playbooks.
A working repo-native harness POC demonstrates repeatable benchmark execution and quality gates.
The committee has sufficient evidence to ratify the standard and decide whether to adopt or build the long-term harness.