Deloitte

From Chaos to Control - A governance model for autonomous agents

Description of the company

At Deloitte, our Purpose is to make an impact that matters for our clients, our people, and society. This is the lens for which our global strategy is set. It unites Deloitte professionals across geographies, businesses, and skills. It makes us better at what we do and how we do it. It enables us to deliver on our promises to stakeholders, while creating the lasting impact we seek.
Harnessing the talent of over 470,000 people located across more than 150 countries and territories, our size and scale puts us in a unique position to help change the world for the better—by bringing together the services we provide, the societal investments we make, and the collaborations we advance through our ecosystems.

Situation

Our customer uses several types of agents to automate processes, workflows, or personal productivity. Anyone at the client can propose their ideas on building an AI Agent for individual problems. The so-called Agent Innovators are a specialized team to help realize these ideas and build an agent for everyone to use.

Problem

1. The Ideal State
In an efficient enterprise environment, AI agents—ranging from autonomous process bots to integrated software workflows—should be governed through a centralized framework. Ideally, organizations should have full visibility into the deployment, platform origin (e.g., Google Vertex AI, Microsoft Copilot, n8n), and functional scope of all agents to ensure security, compliance, and resource optimization.

2. The Problem (The "Gap")
Currently, the ease of deploying AI agents across fragmented ecosystems—including standalone frameworks, low-code platforms, and embedded enterprise software (SAP, ServiceNow)—has led to the emergence of "Shadow AI." Agents are being built and deployed in silos without central oversight. This fragmentation creates significant risks:

• Security Vulnerabilities: Unmonitored agents may handle sensitive data without adhering to corporate security protocols.
• Redundancy: Different departments may build overlapping agents on different platforms, leading to wasted licensing costs.
• Maintenance Debt: Without a central repository, it is impossible to track which agents are active, who owns them, or when they require updates.

3. The Consequence & Research Goal
If left unaddressed, this lack of transparency will lead to an unmanageable "Agent Sprawl," increasing the attack surface for cyber threats and diminishing the ROI of AI investments. This projects aims to design a centralized repository model or governance framework that allows companies to catalog, monitor, and manage heterogeneous AI agents across diverse platforms.

Aims of the project

The goal of this project is to develop a lightweight Governance concept for autonomous AI Agents. This should enhance the responsibility in use of AI Agents and simplify access

• Developing a Governance-Concept to deal with autonomous agents
  • With help of existing frameworks
• Defining a Lifecycle-Modell
• Blueprint of an Agent-Registry and Agent Cards
• Developing a simple prototype of an Agent Registry and Agent Cards

Scopes

• Developing a Blueprint that is presented in Management-Slides
• Providing documentation of the blueprint and prototype
• Developing a prototype using a Google AI Studio

Target group (students)

Skills:

• Basic knowledge of agentic AI
• Analytical and conceptual thinking
• Will to dive into agile project work
• Programming Skills (every level)

Prior experience:

• None

Dates
Please save these dates: Fishing for Experience Termine

Registration
You can apply for Fishing for Experience online.