Adobe · posted · checked
Staff Engineer, Agentic Governance
AWSAzureJavaKubernetesPythonOn-site · BucharestStaff+
Who can apply
On-site · Bucharest
Location: Bucharest
The Opportunity
We are looking for a Staff Engineer to lead the technical direction of Agentic Governance in Adobe's Experience Platform team in Bucharest.
Trust and Governance builds the services that enable Adobe Experience Platform customers to trust us with their data: access control, data usage policies and lineage enforcement, catalog, privacy and consent, audit and security capabilities such as customer-managed keys. Agentic Governance extends these foundations to AI agents: who an agent is, what it may do, and a verifiable record of what it did.
Enterprises are putting AI agents to work on their data, and AI regulation is setting the standard for governing them. This domain is new and owned from Bucharest: you will set how agents are identified, authorized, and audited across Adobe Experience Platform, with patterns other Adobe teams adopt.
What You'll Do
Design and build governance, security and privacy services, integrating with existing technical services.
Drive AI-first architecture across teams: define how agents, tools and MCP servers are authenticated, scoped and observed.
Be the domain's technical leader: lead design, architecture, security, and threat-model reviews across business units, set the test strategy and code standards, and decide on ambiguous problems.
Use AI-assisted development responsibly, and raise the team's practice with evals, guardrails, and quality gates for the agentic components we ship.
Mentor engineers and help shape hiring.
What You Need to Succeed
10+ years of professional software engineering experience, including technical leadership across teams.
Deep, hands-on expertise in identity and authorization: OAuth 2.0/OIDC, RBAC/ABAC, and policy engines (e.g. OPA, Cedar, or comparable). Applying them to workload or agent identity is a strong plus.
Practical knowledge of agentic systems in production, and can explain trade-offs for tools, guardrails, latency, costs and failure modes.