Staff Software Engineer — Content ToolsContent ToolsCoCounsel Agent ContextOverview of the Role The CoCounsel Agent Context team is seeking a Staff Software Engineer to set the technical direction for Content Tools — the purpose-built APIs that give AI agents exact, governed access to Thomson Reuters’ proprietary legal content across various international jurisdictions. Content Tools are the bridge between TR’s authoritative content and the AI products, such as CoCounsel Legal, that depend on them, and they carry high stakes — every result an agent returns must be accurate and grounded in TR’s authoritative content. This is a high-leverage role at the intersection of API design, information retrieval, and agentic systems. You ship to production constantly, and you treat delivery friction as an engineering problem to solve, not a fact of life. You use AI-assisted development as a primary tool, with the majority of code written with AI assistance. And you practice verified spec-driven development: acceptance criteria and an evaluation suite define a tool’s behavior before the implementation exists and gate every change in CI. Because AI writes most of the code, what you truly own is the spec and the eval that verifies it — the eval is the spec, and you are accountable for the outcome, not the lines. The successful candidate brings platform thinking, deep API and evaluation instincts, enterprise-grade security awareness, and a product-minded approach to developer experience — treating each tool’s contract as a product in its own right.
Content Tool Architecture & API Ownership Set the technical direction for the Content Tools catalog: define the design patterns, schema templates, and contribution standards every tool is built against,



and the guidance on when to create a new tool versus extend an existing one — resolving problems of broad scope with little precedent to guide solutionsDrive an API-first, contract-driven approach: every tool ships with a versioned contract, schema, and ownership metadata that makes it discoverable and self-service for consuming teams and agent developers, who should never need to ask you how to use your toolArchitect the retrieval and search layer behind the catalog — relevance and ranking quality on the team’s Vespa based search platform, and the trade-offs between structured and semantic retrieval — partnering with TR Labs on the underlying search algorithmsDefine the agent invocation model for the catalog: clean tool boundaries exposed over MCP (Model Context Protocol), result payloads designed for token economy, and latency budgets that hold across multi-hop retrieval chains, giving agent orchestrators predictable, well-documented interfaces to TR and third‑party contentBuild and own the shared development and testing infrastructure — common libraries, scaffolding generators, mock services, and local environments, built on the team’s Python, FastAPI, and Claude Agent SDK stack — that makes it easy for any team member to build a Content Tool correctly the first timeVerified Spec-Driven Development & Evaluation Own verified spec-driven development as a team-wide standard: define how acceptance criteria and evaluation datasets are written before implementation to specify a tool’s behavior,



IR metrics (precision, recall, NDCG), and CI gates that verify every change and set the bar every engineer’s eval suite followsTreat evaluation as foundational infrastructure the team is still maturing: assess current coverage across tool families, close the highest‑risk gaps first, and prevent teams from redundantly reinventing evaluation capabilityDefine how offline grading pipelines and online signals combine into a continuous, trustworthy quality picture for every tool in productionReliability & Operations Take full operational responsibility for what you and your area own — you built it, you own it, you run it: own the on‑call model and SLO targets for your services, and build the automation that reduces operational toil for the whole teamLead incident response for high‑severity issues, write thorough post‑mortems, and drive systematic fixes that prevent recurrenceBuild the observability that makes reliability visible — including trace‑level instrumentation for agentic retrieval paths, where standard request/response logging cannot explain a non‑deterministic failureOwn performance and cost strategy for the tool layer: capacity planning, caching, and rate limiting that keep the platform reliable and economical as tenants, content volumes, and query traffic grow, with fast, safe delivery treated as a first‑class engineering goal rather than a trade‑off against reliabilityTechnical Leadership & Governance Function as the technical authority other engineers and partner teams turn to for Content Tools API, evaluation, and architecture decisions; author cross‑team Architecture Decision Records for cross‑cutting choices and contribute to the code‑ and design‑review policy the team

📌 Staff Software Development Engineer (Winnipeg)
🏢 Thomson Reuters
📍 Winnipeg