Just as business leaders are beginning to scale generative AI, an even more advanced capability is emerging: agentic AI. AI agents represent a leap forward, enabling organisations to automate complex processes, foster human/AI collaboration and unlock unprecedented business value.
AI agents bring the agency that AI systems need. They blend advanced reasoning from LLMs with planning, orchestration, knowledge, data mining, guardrails and governance to achieve organisational goals through meaningful action.
They make real-time decisions, adapt to new situations and learn from their interactions and reinforced feedback – all under appropriate levels of human oversight. The range of real-world applications is vast, from task-oriented agents that extract and compare data or documents against standards, to fully autonomous agents capable of transforming entire value chains.
The KPMG TACO Framework™ – choosing the right agent for the task
Identifying where and when to use specific AI agents can be complex. KPMG has developed a structured classification system called the TACO Framework – Taskers, Automators, Collaborators and Orchestrators – that helps organisations make sense of the evolving agentic AI landscape and how to leverage its full capabilities.
- TACO
- Taskers
- Automators
- Collaborators
- Orchestrators
- Matrix
Taskers
Taskers execute well-defined singular goals that feature repetitive tasks and minimal complexity, such as reviewing invoices or screening for compliance risks.
They rely on foundational data that can be used across multiple applications and business processes. The ‘human in the loop’ supplies detailed instructions outlining the steps and actions the agent must execute.
Example Tasker applications
- Customer service: Categorise incoming support tickets based on keywords and assign to the appropriate team.
- Vendor due diligence: Screen vendor names against global sanction lists and flag potential issues.
- Invoice validation: Extract and verify data against compliance rules.
- Legal affairs: Convert past legal rulings into a standardised format.
- Administrative: Review employee expense reports for policy compliance and supporting documentation.
Automators
Automators deconstruct more complex goals and manage tasks that span multi-system workflows that can comprise applications, systems, and functional areas.
They draw from the ‘tacit knowledge’ of experienced workers from within the organisation. Automators execute end-to-end processes, orchestrate related tasks, manage dependencies between tasks, and dynamically engage tools, APIs, or MCP servers, as needed.
Example Automator applications
- Supply chain: Streamline workflows by integrating order and transportation management systems, as well as supplier APIs.
- Healthcare: Automate claim adjudication by integrating hospital billing systems, payer systems and regulatory databases.
- Finance: Accelerate month-end close by managing accruals, reconciliations and journal entries across multiple systems.
- E2E order-to-cash: Automate invoice generation, payment reconciliation and reminders.
- Customer onboarding: Adapt dynamically to missing info or compliance checks.
Collaborators
Collaborators function as adaptive AI teammates, working interactively with humans to achieve multidimensional goals, while responding to user feedback and evolving contexts.
These agents integrate multiple AI capabilities (machine learning, NLP, speech, computer vision) and can work with both structured and unstructured data to provide expertise-driven support and augmentation.
Unlike simpler agents, Collaborators maintain contextual awareness and long-term memory across extended complex interactions, enabling them to function as true digital colleagues rather than passive tools.
Example Collaborator applications
- Product development: Synthesize user feedback, suggesting feature prioritisation and generating release plans with timelines and resource estimates.
- Marketing: Brainstorm campaign ideas with marketing teams, refine ad copy based on style preferences and automatically schedule social media posts.
- Financial planning and analysis: Work alongside analysts to develop scenario models, evaluate assumptions and provide insights based on historical data patterns, while adapting to the specific analytical approach preferred by each team member.
Orchestrators
Orchestrators are an advanced class of agentic system designed to coordinate multiple agents, tools, and interdependent workflows at scale. These agents operate as intelligent control towers, dynamically selecting from a broad ecosystem of specialised agents and services to achieve complex, cross-functional objectives. They rely on explicit knowledge, which is more objective, quantifiable, and often technical.
Operating in real time, Orchestrators enable dynamic resource optimisation, multi-agent choreography, task delegation, inter-agent communication (leveraging protocols such as the agent-to-agent (A2A) protocol) and the reconciliation of dependencies across systems.
Preparing for their adoption requires addressing the technology stack and codifying robust Trusted AI protocols.
Example Orchestrator applications
- Finance: Manage inter-company reconciliation processes by orchestrating agents for currency conversion, account matching and exception reporting.
- Regulatory compliance: Orchestrate multi-jurisdictional compliance monitoring by assigning agents to review local regulations, ensure adherence and escalate violations for manual review.
- Multi-entity procurement optimisation: Orchestrate sourcing, negotiation and compliance agents across global subsidiaries, within and outside of the enterprise.
Choosing the right agent for the task
Agentic AI advantage: Unlocking next-level value
Download our global report to explore strategies for integrating agentic AI.
Get in touch
From experimentation to implementation, our specialists are here to guide your agentic AI journey.
Levi Watters
Head of Digital Build Services & National Sector Lead, Telecommunications
KPMG Australia
- Levi Watters
- Dhawal Jaggi
- Brad Daffy
- Lisa Jenkinson
Let’s talk AI – book a discovery workshop
Ready to explore what AI can do for your business?
Whether you’re starting small or scaling big, we’re here to help you move forward with clarity, confidence, and impact. Fill out the form and a KPMG AI specialist will be in touch.
First in the world to be ISO certified for AI by BSI
KPMG Australia proudly leads the charge in AI excellence, as the first firm globally to receive ISO 42001 certification for AI Management Systems by the British Standards Institution (BSI).
It’s another way we’re showing our deep commitment and dedication to Trusted AI.
How KPMG can help
The KPMG TACO Framework™ is a structured approach that helps organisations unlock value from agentic AI while managing risks and enabling trust.
KPMG offers a portfolio of services tailored to help you overcome key barriers to agentic AI adoption. By optimising the collaboration between what humans and agents can do, we help you leverage their full potential to create lasting value.
