AI Operations Developer

We need a high leverage builder who can turn delivery and operational processes into AI enabled scalable internal systems. This role sits between product management delivery leadership enterprise data and AI tooling - designing solutions that are secure cost aware and production ready not experimental demos.

Core Responsibilities

• AI-enabled operations design
• Design AI workflows for delivery PMO resourcing reporting and governance.
• Focus on boring high-friction internal processes (status reporting intake metadata correlation risk surfacing).
• Explicitly control AI scope token usage and cost profiles (e.g. 1anddone passes vs. conversational loops).
• Data access & integration
• Build and/or define secure delegated access to enterprise data.
• Microsoft Graph (mail chats meetings transcripts) SharePoint/OneDrive Salesforce ServiceNow Dataverse.
• Create custom connectors or lightweight services so AI tools can access data safely and consistently.
• Package complexity behind simple actions (e.g. Search GetMeetingTranscripts GetProjectRoles).
• Optimize for reuse across multiple agents and teams.
• Human-in-the-loop system design
• Design workflows that assume AI is probabilistic and humans review override and approve.
• Use adaptive cards approvals and structured outputs rather than free-text magic.
• MVP-to-production translation
• Identify why most AI pilots die: licensing permissions performance and governance.
• Proactively design guardrails before scale becomes a problem.
• Adoption & enablement
• Ensure tools built are used-track adoption gather feedback and iterate.
• Train and enable stakeholders and the organization to use AI-enabled workflows and communicate clearly to non-technical stakeholders what the tools do and dont do.

Required Skills

• Deep expertise across the Microsoft Power Platform (Power Automate Power Apps Copilot Studio) and Microsoft Graph API with a clear understanding of delegated vs. application permissions and their security consent and operational implications.
• Hands-on experience working with meeting transcripts chat data email and enterprise metadata including building custom connectors and designing appropriate authentication models scopes and consent flows.
• Proven experience building complex AI agents in Copilot Studio (not just prompt authoring); understands token economics context limits and cost-amplification risks and has intentionally constrained AI behavior to reduce cost risk and unintended outcomes.
• Operates with a privacy-first mindset assuming every data path is a liability until proven otherwise and consistently designing for minimum necessary access rather than maximum convenience.
• Strong awareness of platform limitations-knows what Microsoft Graph Copilot Studio and the Power Platform cannot do and designs around those constraints intentionally instead of discovering them in production.
• Pragmatic polyglot builder who writes whatever the platform requires (C# Python Power Fx JSON YAML) without relying on a dedicated development team to unblock progress.
• Fluent in IT and identity infrastructure; capable of authoring permission requests validating access and navigating Azure AD / Entra ID as a peer to IT Operations rather than a dependent.
• Operates beyond documentation by testing edge cases validating assumptions and documenting findings when official guidance ends.
• Comfortable engaging architects and executives in the same conversation translating technical constraints into clear business and risk tradeoffs.
• Brings a strong builder mentality-cannot stop at theory and is driven to make solutions work in production.
• Demonstrates antifragile curiosity leaning into permissions APIs and constraints rather than avoiding them.
• Acts as a force multiplier by enabling others and sharing solutions rather than hoarding knowledge.
• Maintains a healthy skepticism of AI hype clearly understanding where AI fails and designing systems that account for those limitations.
• Comfortable pushing back when architectural security or cost decisions do not make sense.