Design, build, deploy, and support core components of the harness - the guides, feedback loops, guardrails, and shared context that turn raw model capability into production-grade engineering. This is a hands-on role focused on systems and leverage, not hand-writing application code
Own and evolve feedforward guides - agent instruction files, reusable skills, architectural rules, reference docs, and codemods - and drive team-wide standardisation so agents get it right the first time
Build feedback sensors - custom linters, static analysis, structural and architecture-fitness tests, verification loops, and LLM-as-judge reviewers - that catch issues automatically before they reach human reviewers
Own quality gating and release criteria for agent-produced work, defining authority boundaries for what agents may merge unaided and the escalation rules for what must route to a human
Establish LLM testing infrastructure and evaluation approaches that ensure AI-generated output meets quality and safety thresholds; apply consumer/contract testing (e.g. Pact) where service integration reliability matters
Run the steering loop - when an agent repeats a mistake, engineer a control so it can't happen again - and treat repository knowledge (docs, specs, context) as the system of record, fighting drift with continuous garbage collection
Decide where each control runs in the path to production - fast checks pre-commit, more expensive checks post-integration, and continuous sensors that scan for drift outside the change lifecycle - keeping quality as far left as is economical
Improve observability into agent work and track the measures that matter - cost per merged PR, time-to-merge for agent-assisted PRs, review velocity relative to PR size, defect escape rate, and agent-PR survival rate - using them to decide where to invest next
Partner with product and platform teams to turn specifications and acceptance criteria into enforceable controls
Serve as a source of technical expertise and mentor engineers across teams in harness practices and the effective, responsible use of AI tools