AI Infrastructure Engineer

Tooling in AI Platforms and Systems are replaced faster than anyone can be trained on it, so a hire made on product familiarity is a hire with a short shelf life. This profile therefore sets out the qualities that do not expire and cannot be taught on the job, followed by the technology landscape the role is expected to operate in and the key responsabilities.

What the person must be like

- Learns from primary sources. Goes to the original documentation and release notes rather than to summaries or hearsay, and can tell the difference between understanding a system and recognising its vocabulary.
- Thinks in boundaries. Asks what a system can do, on whose authority, with which data, and who sees the result, before asking how to switch it on. Treats the joins between systems as where risk actually lives.
- Decides under uncertainty, and revisits. This field moves faster than the evidence about it. Decides on incomplete information, records the assumptions and what would change them, and reverses the decision later without treating that as a defeat.
- Sceptical by default, including of AI output. Treats fluent, confident output from a vendor, a colleague, or a model — as a claim to be verified rather than a fact, and applies the same standard to their own conclusions.
- Governs in order to enable. Designs the lightest set of controls that makes the business goal safe. Knows that a control people route around is not a control, and that blanket prohibition produces hidden adoption rather than safety.
- Judges risk from observation, not only from checklists. Forms an independent view of exposure from usage data, logs, incidents, and what people are actually doing, instead of waiting for a framework to declare something a risk.
- Writes so that others can act. Governance that exists only in one person's head is not governance. Writes decisions and procedures a colleague can follow unaided and an auditor can read without a translator.
- Works across teams without authority. Most of what this role needs is delivered by people who do not report to it. Persuades, negotiates, escalates proportionately, and can say no in a way that survives the meeting.
- Raises the uncomfortable thing early. Surfaces a risk or an error while it is still cheap to correct — including when it is their own, and including upwards.





Technology landscape The platforms below illustrate the landscape this role governs. Candidates will typically have worked with two or three; none is expected to have worked with all, and expertise in any single one is not a requirement. The role is to govern whichever of these the company adopts next — and most of what this person will govern in eighteen months is not on this list yet.
- Enterprise AI assistants. Microsoft 365 Copilot, ChatGPT Enterprise, Claude Enterprise, Devin AI, Google Gemini Enterprise, Perplexity Enterprise, Glean.
- Agent platforms and control planes. Microsoft Agent 365, Copilot Studio, Agent Builder, Salesforce Agentforce.
- Low-code and data platforms. Power Platform, Microsoft Fabric and Copilot Studio.
- Model and agent cloud platforms. Microsoft Foundry, AWS Bedrock, Google Vertex AI.
- Coding and autonomous development agents. GitHub Copilot, Claude Code, OpenAI Codex, Cursor, Devin.
- Identity, security, and compliance tooling. Microsoft Entra, Microsoft Purview, Microsoft Defender.

Key Responsibilities
- Platform Administration: Administer and configure the corporate AI platform estate at tenant level, including service enablement, environment strategy, admin role delegation, and controlled adoption of preview and general availability releases across Microsoft and non-Microsoft AI platforms.
- Agent Lifecycle Management: Operate the agent registry end to end — discovery and onboarding of agents (including shadow agents), ownership assignment, approval and publication workflows, certification, change control, and retirement — across Copilot Studio, Agent Builder, Power Platform, and third-party agent platforms.
- Identity and Access Requirements: Define the identity and access requirements of each agent and AI platform — intended scope, permissions, consent model, and least-privilege boundaries — and hand them over to the Entra identity team for implementation, then validate that the delivered configuration matches what was requested.




- Guardrails and Policy Enforcement: Translate approved corporate AI safeguards into enforceable technical controls and evidence their operation — data loss prevention, connector and environment boundaries, tenant isolation, admin consent, sensitivity labelling, and audit logging — in partnership with Security, Privacy, and Compliance.
- Automation: Automate provisioning, governance, remediation, and reporting using PowerShell 7, Microsoft Graph, and platform administration APIs and CLIs, with version-controlled, peer-reviewed, and documented code.
- Service Operations: Act as escalation point for AI platform incidents and service requests; define runbooks, monitoring, and service levels; and drive problem management through to root cause and permanent fix.
- Collaboration and Enablement: Partner with the Entra identity team, the AI Centre of Excellence, Security, Legal and Privacy, Data, and business units to onboard recent agent use cases safely, and coach makers and agent owners on governance requirements and platform boundaries.
- Cost and Licence Management: Manage licensing, capacity, and consumption across the AI estate; deliver segregated cost allocation and chargeback reporting per platform and business unit; and identify optimisation and rationalisation opportunities.
- Usage Analytics and Adoption Reporting: Build and maintain dashboards covering agent inventory, usage, value, and risk signals, and report adoption, service health, and return on investment to platform owners and business stakeholders.
- Release Monitoring and Service Incorporation: Review the continuous flow of AI feature announcements across the platforms in scope — the Microsoft 365 Message center and roadmap, Power Platform, Copilot and Fabric release notes, and the updates published by OpenAI and Anthropic — assess the impact of each relevant change on configuration, governance controls, licensing, cost, and users, and turn that assessment into a documented decision to adopt, contain, or defer, followed by the pilots, policy updates, and communications that the decision requires.
- Documentation and Standards: Maintain governance standards, administration runbooks, architecture decision records, and maker-facing guidance that keep the platform supportable, transferable, and audit-ready.

📌 AI Infrastructure Engineer (Bengaluru)
🏢 Carnation Infotech
📍 Bengaluru