There is a new form of AI for professionals that promises to handle complex tasks. It is called agentic AI. Here is what you need to know about AI agents.

Agentic AI systems operate autonomously to resolve complex problems. Conversely, generative AI (GenAI) responds to a specific prompt from a user to generate an output.

GenAI is designed to be assistive. In comparison, agentic AI has been developed to function autonomously and execute workflows. GenAI chatbots require direct human prompting to produce output. Agentic AI agents offer the potential to execute multi step processes without direct human supervision.

The transformational impact of GenAI on manual workflow processes has dominated the spotlight in recent years. The rise of AI agents and their autonomous capabilities can boost professionals’ productivity as well. Read on to learn what agentic AI can do for professionals.

How does ‘agentic AI’ work?

Agentic AI systems work by “mimicking human decision-making” according to Vincent Koc, Program Mentor, Applied Data Science Program, at Massachusetts Institute of Technology, and Lecturer at UNSW.

Vincent has over two decades of experience as an engineering technologist. Vincent is a fellow at the Institute of Managers and Leaders Australia. He shares his expertise in generative AI, data, and governance at universities here, and in the U.S.

“Like human agency with agentic AI systems, the ideas are to undertake complex actions with minimal supervision, making decisions based on predefined goals or learned patterns in a human-like manner,

“Generative AI and other forms of models might be components of the core part and the model that becomes the agent.

“Most generative AI models could be made agentic in the right context, framing and tooling given to the model to carry out given actions within a self-directed and self-aware manner,” said Vincent.

Agentic AI’s multi-step problem solving

Agentic AI systems solve multi-step problems by applying four key principles: perceive, reason, act and learn.

While agentic AI is autonomous, the AI model operates within set parameters. Agentic AI uses a process of reasoning and iterative planning based on several trusted data sources. The artificial intelligence model independently identifies challenges, and analyses information to provide solutions to solve multistep problems.

Agentic AI systems evaluate information collected from specific data sources to produce actions. The agentic AI system creates a draft roadmap to solve a complex query. Autonomous AI agents can execute work individually or self-organise as a team of AI agents to tackle a query.

How agentic AI solves problems:

• Perceives: Gathers and processes data from various sources within trusted data databases.
• Reasons: Analyses and understands tasks, generates solutions, coordinates to complete tasks like content creation, or preparing recommendations.
• Acts: Agentic AI systems can execute tasks based on the plans it has formulated using integrated software and external tools.
• Learns: An agentic AI system will adapt from new interactions and observations via a feedback loop, or “data flywheel.”

“Over time, AI agents learn and improve by creating a data flywheel, where data generated through interactions is fed back into the system, refining models and increasing their effectiveness,” Vincent explained.

How could agentic AI help with professional workflows?

Agentic AI’s automation capability could be applied to a series of tasks involved in professional workflows. These may include case management, client management or billing management. Agentic AI systems could help professionals work more efficiently by converting data into insight.

“The million-dollar question is always not about the technology but what true value and bottom-line impact could this bring to any organisation,” said Vincent.

“This ties to the concept of defining critical ‘use-cases’ which can be assigned to driving value through revenue impact, cost reduction or both.

“When we consider what agentic systems can support with autonomous problem solving and acting whilst augmenting and complementing existing human resources in an organisation there can be considerable impact if executed onto the right problems.”

Incremental implementation of agentic AI is key

Incorporating an agentic AI system into business or government operations should begin with a strategic approach, according to Vincent. Problem mapping and incremental implementation is key in this process. Simply defining the problem and the use-cases is “just half of the solution.”

“Agentic AI systems can significantly enhance business and government operations by automating decision-making processes, optimising resource allocation, and improving strategic planning in a broad range of applications,” said Vincent.

It is best practice for humans to be kept in the loop of operation due to agentic AI’s autonomy. Human oversight and expertise are essential to fact-check and manage the quality of deliverables.

“Taking a systematic approach to how you define the right use-cases with an internal score card and compass will help you navigate through the change a lot better,” said Vincent.

“This process ensures that AI initiatives align with organisational goals and deliver measurable value efficiently.”

RAG model can be a useful guardrail

The Retrieval-Augmented-Generation model (RAG) is one of many techniques that agentic AI can use as a guardrail.

Professional-grade, artificial intelligence models that operate using the RAG model only source from data provided and cannot go beyond these parameters. The technique enhances accuracy and reliability by removing ambiguity in user queries.

“When it comes to ethical considerations, the AI system regardless of its ability to be self-aware or self directed will still require some explicit ethical programming versus intrinsic ethical constructs that a human would comprehend through societal norms and cultural exposure,” Vincent added.

Thomson Reuters’ Tech, AI and the Law 2024 report revealed 31% of private practice legal professionals use unofficial AI for work. Legal leaders will now need to consider providing staff with access to trusted and ethical AI.

“We also need to deliver on the impact whilst managing the change and ethical considerations that come with this,” said Vincent.

AI agents can accelerate productivity

Grounding agentic AI systems to address specific organisational needs is a clear opportunity to speed-up productivity.

Agentic AI systems can direct agents to undertake actions that extend beyond a GenAI assistant’s aptitude. Agentic AI systems have the capability to assemble a team of agents to work collaboratively. An agentic AI system can direct single agents execute project specific tasks solo.

Based on the Tech, AI and the Law 2024 report data, 95% of law firm professionals believe “AI is no substitute for thorough legal work, but it helps to accelerate it”. AI agents will likely become more commonplace in the legal, tax and finance professions. In October 2024, Thomson Reuters acquired Materia, an AI agent purpose-built for the tax, audit and accounting industry.

How does Materia automate workflows?

Materia’s agentic AI autonomous agents have been supporting accountants since 2022. According to David Wong, Thomson Reuters Chief Product Officer, agentic AI can be applied to a range of professional services.

“Our vision is to provide each professional we serve with a Gen AI assistant. Materia will further accelerate this vision for our tax, audit and accounting customers.”

“Once fully integrated, Materia will transform work and unify the entire customer experience with applications across our tax, audit, and accounting portfolio,” said David.

“We are excited by the potential of combining Materia with Thomson Reuters content, know-how, and solutions.”

Guided by objectives, Materia dispatches its autonomous agentic AI agents to deconstruct problems into individual steps. By retrieving the most relevant information, agentic AI delivers value to their clients by improving efficiency and effectiveness.

By eliminating low-value tedious tasks, according to Kevin Merlini, CEO of Materia, accounting teams can invest time in higher value advisory work with increased quality.

“We will be able to move more quickly toward this objective and will unlock new opportunities to deliver highly requested solutions for our customers,” Kevin concluded.