AN ENTERPRISE AGENT MANIFESTO

Written from the large enterprise perspective and personal decades of experience, this manifesto provides basic guidance on implementing agent technology. The hypothesis is that agents will emerge in a predictable way. While timing remains unclear, we offer a structured framework for navigating this evolution.

What is Agent Orchestration?

Agent orchestration mirrors proven organizational structures, coordinating multiple specialized AI agents to collaborate on complex tasks. Just as human teams require coordination and clear responsibilities, AI agents need orchestration to work together effectively and efficiently.

The Finance Team Example

Consider the Accounts Payable function as an agile team of agents, each with a specific persona:

• Finance Manager Agent: Prioritizes payment activities and ensures alignment with cash flow objectives and business policies
• Compliance Lead Agent: Makes procedural decisions and guides implementation of controls, ensuring adherence to regulatory requirements and internal policies
• Transaction Processing Agents: Process invoices, reconcile statements, and execute payments based on established approval workflows
• Audit/Validation Agents: Ensure accuracy through automated three-way matching, duplicate detection, and fraud pattern recognition
• Led by a Process Orchestration Agent: Coordinates the end-to-end workflow, identifies bottlenecks, and escalates exceptions requiring human intervention

This model assumes architecture and vision established at a higher enterprise level to ensure strategic alignment and that technical debt is minimized.

The power of this approach is that it leverages established team structures that organizations already understand, making adoption more intuitive and granular.

Agent Taxonomy: A Framework for Enterprise Implementation

As organizations develop their agent strategy, they must consider a spectrum of agent types based on function, scope, and autonomy. This framework provides a structured approach to agent classification:

By Function

1. Information Agents
2. Task Automation Agents
3. Decision Support Agents
4. Autonomous Agents

By Scope

1. Personal Productivity Agents
2. Team Collaboration Agents
3. Departmental Agents
4. Enterprise-wide Agents

By Autonomy Level

1. Level 1: Assistive
2. Level 2: Collaborative
3. Level 3: Supervised Autonomy
4. Level 4: Full Autonomy

This multi-dimensional framework allows enterprises to map their agent strategy according to organizational readiness, risk tolerance, and business objectives. Most organizations should begin with high-function, narrow-scope, low-autonomy agents before progressing to more sophisticated implementations that require robust governance and oversight mechanisms.

The Evolution of Enterprise Agents

The requirements for agent interoperability may grow very quickly. Within the enterprise, adoption will likely follow this progression:

• Domain Agents: Specialized for functional areas like Accounting, HR, Supply Chain functions. These agents excel in narrow domains with deep expertise but limited cross-functional capabilities. For example, an HR agent might automate onboarding processes while having no visibility into financial systems.
• Cross-Domain Agents: Working across functional boundaries to coordinate activities and information flow between specialized domain agents. These more sophisticated agents will emerge as organizations recognize the limitations of siloed implementations and seek greater integration for end-to-end process automation.
• Externally-facing Agents: Engaging with customers, partners, and suppliers while representing the enterprise’s interests and brand. These will be the most complex to implement as they must balance customer experience with security, compliance, and brand representation.

This natural progression mimics how enterprise software has evolved historically—from function-specific tools to integrated systems to customer-facing platforms.

Supplier Landscape

So how do organizations implement this kind of complex ecosystem?

Let’s do a quick review of the options.

Vendors

• CRM platforms: Will embed agents for sales automation, customer service, and relationship management
• ITSM solutions: Will incorporate agents for incident management, service requests, and infrastructure monitoring
• ERP systems: Will integrate agents for process automation, compliance, and operational optimization

Each will incorporate agentic technology but brings risks of technical lock-in, extortionary pricing strategies, or other disruptive changes due to M&A. Enterprises should be particularly cautious about proprietary agent frameworks that don’t offer portability or interoperability with other systems.

Open-source

Open-source alternatives allow for greater flexibility and control but require more internal expertise to implement and maintain. These solutions offer a hedge against vendor lock-in and can serve as the foundation for truly organization-specific agent capabilities. However, they typically require stronger internal technical capabilities and longer implementation timelines to realize their full potential.

Hypotheses about Governance

Each agent needs to be auditable, with comprehensive logging of actions, decisions, and data accessed. Without this capability, enterprises face significant compliance, security, and operational risks as agent usage scales.

An Agent Control Plane or a similar approach is necessary to standardize agent interactions with full visibility. This centralized governance layer should provide authentication, authorization, monitoring, and policy enforcement across all agent activities. Think of it as air traffic control for your enterprise agents—ensuring safe, coordinated operations within defined parameters.

For the Enterprise Customer: What Do You Do?

Enterprise customers face a rapidly evolving landscape of agent technologies across vendor ecosystems and open-source alternatives. The strategic approach should be threefold:

1. Start with pilot programs in contained domains where ROI can be clearly measured. Focus on use cases with well-defined processes and clear metrics for success. This approach limits risk while providing tangible evidence of value to support broader adoption.
2. Establish governance frameworks early, including auditing capabilities and an agent control plane. Implementing governance after widespread adoption is significantly more difficult and expensive. Early investment in these capabilities will pay dividends as agent usage grows.
3. Prioritize interoperability standards in all vendor selections to prevent lock-in. Demand APIs, standard data formats, and portability commitments from vendors to ensure your agent ecosystem remains flexible as technologies evolve.

The key is balancing immediate productivity gains from purpose-built domain agents while building toward a coherent multi-agent architecture that preserves enterprise autonomy and cost control. This requires thinking both tactically (which processes can benefit from agents now) and strategically (how will these agents work together in the future).

Conclusion: The Enterprise Agent Investment Thesis

The evolution of enterprise agents represents both tremendous opportunity and significant risk. Organizations that approach agent adoption strategically will gain sustainable competitive advantages through enhanced productivity, reduced operational costs, and improved decision-making capabilities.

Here are some hypothetical outcomes.

Curated from Claude-3.7-Sonnet

Organizations that establish clear governance models, prioritize interoperability, and balance vendor solutions with open-source alternatives will be best positioned to capture these returns while managing the inherent risks of this transformative technology.