By Swami Chandrasekaran, KPMG.
AI agentic systems represent a transformational shift in enterprise AI, moving beyond passive information retrieval to proactive execution and decision-making. Agentic systems don’t just retrieve and answer queries – they take actions using tools.
They enable autonomous planning, reasoning, and action-taking, dynamically adapting to evolving business objectives. With human oversight (both in and on the loop), AI agents act as collaborative partners, executing workflows, making real-time decisions, and driving business outcomes.
Unlike simpler generative models, AI Agents can orchestrate intricate tasks, manage cross-application workflows, and collaborative interactions between other AI agents and human operators.
Understanding AI Agentic Systems Through the T.A.C.O. Framework
As AI Agentic Systems evolve, a structured framework is essential to understand and categorize them based on their capabilities. We can look and assess them by examining the complexity of the goals they fulfil, the depth of planning required, and the level of coordination & orchestration involved.
To make sense of these variations, I propose the TACO framework – which classifies agents into four key types: Taskers, Automators, Collaborators, and Orchestrators. Each of the type leverages the same foundational tools and capabilities – goal interpretation, reasoning engines (using advanced models including LLMs), memory, tools, and orchestration – but differs in goal planning, execution and complexity.
The Four Main Types of AI Agent
Taskers focus on accomplishing singular goals by breaking them into structured, repeatable tasks, making them easy to deploy, scale, and monitor. Examples include:
- Invoice Validation Agent – Extracts and verifies financial data against compliance rules to reduce manual errors.
- Journal Entry Verification Agent – Ensures accounting accuracy by checking posting completeness and flagging anomalies.
- KYC Compliance Agent – Screens customer onboarding documents, detecting compliance risks in regulatory watchlists.
Automators handle goals that require seamless integration across multiple enterprise systems, automating end-to-end business processes and cross-functional workflows. Think procure-to-pay, order-to-cash, and compliance auditing. Examples include:
- Supplier Risk Assessment Agent – Evaluates supplier risks by integrating external risk databases with internal compliance, procurement, and ERP systems.
- Healthcare Claims Processing Agent – Manages patient insurance benefits validation, claims adjudication, and payout processing, reducing manual workload and errors.
Collaborators act as AI teammates, working contextually & closely with human operators, enhancing productivity by learning from interactions and refining recommendations. Examples include,
- Legal Contract Analyzer: Reviews contracts, flags high-risk clauses, and suggests modifications based on corporate policy while working interactively with legal experts and teams
- Investor Relations: Augment and assist in preparing earnings reports, analyzes shareholder concerns, and coaches as well as suggest talking points for investor calls.
Orchestrators involve multi-agent ecosystems where agents collaborate not only with humans but also coordinate with other AI agents to achieve complex tasks at scale while dynamically adapting to real-time changes. Examples include,
- Cross-border Compliance Monitoring
- Capital Project & Investment Program Coordination
Scaling AI Agents: From Experiments to Enterprise
AI agentic systems are evolving rapidly. Enterprises face increasing pressure to keep pace with the changes, advancements and announcements. The TACO framework provides a structured approach to scaling AI from singular goal fulfilment to fully orchestrated, adaptive multi-agent systems that seamlessly integrate into business operations.
However, scaling AI agents successfully requires strategic technology choices—balancing open-source frameworks, commercial platforms, emerging startups, and off-the-shelf solutions to ensure flexibility, interoperability, and long-term impact.
Navigating the Trade-Offs. The Balancing Act. Polyglot it is.
This decision hinges on a key trade-off: rapid deployment vs. deep customization. To avoid vendor lock-in and remain flexible as the AI landscape matures, enterprises should adopt a polyglot approach – leveraging multiple agent frameworks, platforms, and tools to ensure adaptability as technologies evolve and industry standards solidify. However, flexibility alone is not enough. Organizations must simultaneously foster grassroots innovation while maintaining a structured governance framework. This balance ensures compliance, transparency, and interoperability—allowing AI agents to scale securely, build trust, and prevent fragmentation and inefficiencies
As we enter the ‘year of agents,’ organizations must act decisively prioritizing platform selection, data readiness, governance, knowledge elicitation, architectural standardization, and continuous innovation to drive enterprise-wide AI adoption with Trust.
A scalable, well-governed agentic ecosystem will be the foundation for long-term AI success, ensuring compliance, interoperability, and sustained business impact. AI Agents are here –evolving, executing, and transforming enterprises.
–
Many thanks to Swami at KPMG for allowing Artificial Lawyer to publish this article, with some minor changes, and which he first posted on LinkedIn. You can read more about US-based Swami and his work on AI at KPMG here.