SUPRA ITS
Managed IT | Cybersecurity | Cloud | BPO
AI Product Owner
Pumpkin AI Platform
Supra Canada Technologies Ltd.
About Pumpkin AI
Pumpkin AI is an enterprise agentic AI platform built and operated by Supra ITS, part of the Canon group. The platform combines a library of specialist agents with a central control plane: admins create, entitle and deploy agents; end users work with them over documents, data, policy and operational systems.
Recent delivery includes a Dashboards agent, an eLearning agent that converts uploaded training material into video and audio learning content, and a no-code UI agent builder that lets non-developers assemble their own agents. Pumpkin AI runs in the cloud and on-prem, and is positioned to work alongside Microsoft Copilot and other enterprise AI stacks rather than against them.
The role
We are hiring an AI Product Owner to sit with the Pumpkin AI engineering team every day and own the backlog that turns platform capability into enterprise-ready product. You will be the single voice of the product for the squad: writing and ranking stories, defining acceptance criteria, accepting work, and holding the line on quality and compliance.
Your flagship mandate is the governance layer - the controls that make Pumpkin AI safe to sell into regulated, cost-conscious enterprises: spend and cost attribution, security and access control, no-code agent creation in the UI, privacy, and audit compliance. If governance is weak, enterprise deals stall. You own that outcome.
This is a hands-on Scrum Product Owner role, not a roadmap-only position. You will be in refinement, daily scrum, sprint review and retrospective with engineers, and you will be expected to reason about agent behaviour, tool calls, model routing and data boundaries well enough to make real trade-offs in the room.
What You Will Own
- The Pumpkin AI product backlog: epics, stories, acceptance criteria, ranking and release scope, traceable from business outcome to merged code
- The governance layer roadmap across cost, security, custom agent development in the UI, privacy and audit compliance
- Agent lifecycle as a product surface: create, test in sandbox, entitle, deploy, monitor, retire
- The no-code UI agent builder experience - what a non-developer can safely build, and what requires engineering
- Definition of done for agent features, including evaluation criteria, failure modes and fail-closed behaviour
- Stakeholder alignment across Supra ITS service delivery, security, client-facing teams and Canon group stakeholders
- Release notes, demo narrative and the sandbox proof shown to clients and prospects before deployment
The governance layer: what good looks like
- Cost: per-agent, per-tenant and per-user spend visibility, model and token attribution, budgets, quotas and alerting - no surprise invoices
- Security: role-based access control, entitlements, secrets handling, tenant isolation, and agent permissions that are explicit and reviewable
- Custom agent development in the UI: guardrails, approval workflow and promotion path so business users can build agents without creating shadow IT
- Privacy: data residency, PII handling, retention and deletion, prompt and response redaction, and clear rules on what leaves the tenant boundary
- Audit compliance: immutable logging of prompts, tool calls, data access and admin actions, with exports that satisfy SOC 2, ISO 27001 and client audit requests
- Policy as product: controls that are configurable per client, not hard-coded, and demonstrable in a sandbox before a client signs
How You Will Work With The Team
- Run the Scrum cadence with the engineering team: backlog refinement, sprint planning, daily scrum, sprint review and retrospective
- Write stories with testable acceptance criteria; accept or reject work against them without negotiating the bar down at the end of a sprint
- Partner with the Architect on ADRs so product decisions and technical decisions are recorded, not rediscovered
- Work with QA on test coverage for agent behaviour, and with UX on admin and builder workflows
- Keep a sandbox-first discipline: prove behaviour in Admin Sandbox before anything is entitled or deployed to a client
- Balance client-driven requests against platform coherence - say no clearly, and explain the trade-off
Must-have Qualifications
- Proven Product Owner experience in a Scrum team, working with engineers daily - backlog ownership, story writing, acceptance, release scope
- Experience leading or owning the product for an Agentic AI team: agents, tool calling, retrieval, orchestration and evaluation in production, not pilots
- Demonstrable work on an AI governance or platform control plane: cost controls, access control, privacy, logging and audit
- Enough technical depth to debug a conversation with the team: you can read an agent trace, reason about a tool failure, and judge whether a control is real or cosmetic
- Enterprise software judgment: multi-tenancy, entitlements, RBAC, and the difference between a demo and something a client auditor will accept
- Clear written communication: stories, ADR input, release notes and executive updates that stand on their own
- Comfort with ambiguity and with a platform that is still being shaped - you set direction rather than wait for it
- [5+ years] product ownership or product management, including [2+] years on AI, data or platform products
Nice to have
- Hands-on exposure to other agentic platforms: Google (Vertex AI / Agentspace), Microsoft (Copilot Studio, Azure AI Foundry) or AWS (Bedrock Agents) - strongly helpful
- Experience positioning an AI platform alongside Microsoft Copilot rather than in competition with it
- Background in regulated or audited environments: SOC 2, ISO 27001, GDPR, PIPEDA, HIPAA
- Familiarity with RAG, vector stores, model gateways and spend attribution (LiteLLM-class routing)
- MSP or managed services context: ticketing (ConnectWise-class), service delivery, client onboarding
- Experience with no-code or low-code builder products and the support burden they create
- Certified Scrum Product Owner (CSPO), PSPO or equivalent
- On-prem and dual-cloud delivery experience (AWS / GCP / Azure, Docker, Kubernetes)
What This Role Is Not
- A roadmap-and-slides role detached from the engineering team
- A project manager role focused on status reporting rather than product decisions
- A prompt-writing or AI-evangelist role without ownership of the backlog and the governance outcome
- A Scrum Master role - we expect you to partner with one, not replace them
How We Work
Stories to ADRs to TDD implementation, review, then QA. Caps and fail-closed behaviour are product requirements, not engineering preferences. Every capability is proven in Admin Sandbox the same way a client will use it in production, and governance controls ship with the feature they govern - not in a later hardening sprint.
How To Apply
Send your resume and a short note on one agentic AI product you owned: what the agents did, how you governed cost and access, how you measured quality, and what you had to say no to.
Skills: ai governance,agentic ai platforms & autonomous ai agents,scrum,product ownership & product backlog management,ai evaluation & testing
