In 2025, enterprise leaders are no longer debating the relevance of agentic AI â thatâs a given. The focus has shifted to making it work.
Agentic AI systems are capable of memory-based reasoning and autonomous decision-making. Analysis of CEO earnings calls in Q4 2024 found that agentic AI was one of three themes that gained noticeable traction in that period, alongside tariffs and reshoring.
Leading firms are using agentic AI to automate complex tasks, interpret unstructured data, and manage workflows across departments. Gartner expects 15% of routine work decisions to be made autonomously by 2028, up from zero in 2024. It also forecasts that one-third of enterprise software will include agentic AI by the end of this period, up from less than 1%.
But successful deployment will require careful management. CFOs, with their broad oversight and focus on execution, are well-placed to turn experimentation into value.This article explores three key questions: what agentic AI is, how itâs evolving in the enterprise, and what CFOs must get right to make it deliver.
At its core, agentic AI works by breaking a goal into actionable steps. It reasons through options, refines its approach as it learns from past outcomes.
There are three elements to defining agentic AI.
Letâs unpack that definition in more detail.
1-Goal-seeking software with autonomy. In practice, agentic AI behaves more like a goal-seeking collaborator than a traditional tool. Ask it to âplan my weekâ and it can scan your calendar, block time for focused work, book meetings, and adjust the schedule as conflicts arise. It doesnât simply follow a checklist. Instead, it evaluates what needs doing, adapts to changes, and applies the right tools to deliver a result.
At its core, agentic AI works by breaking a goal into actionable steps. It reasons through options, refines its approach as it learns from past outcomes, and uses systems such as browsers, APIs, or spreadsheets to take action. This is not static automation or a reactive chatbot. It is proactive software that operates with intent.
Imagine assigning an agent the task âoptimize our quarterly reporting workflow.â It can:
Think of it as a digital junior controller: fast, tireless, and context aware. When applied to reporting, forecasting, or compliance, agentic AI enables a shift from task management to insight generation.
2-A step beyond automation and generative AI. Unlike traditional automation or even generative AI, agentic AI doesnât wait to be told what to do. It operates with intent: pursuing goals independently, adapting as it goes, and deciding what action to take next.
Where traditional automation follows predefined scripts, agentic AI can handle ambiguity and change. And while generative AI creates outputs when prompted, agentic AI moves with purpose. Itâs not answering questions; itâs solving problems.
Think of it this way: automation is like cruise control; generative AI gives you directions. Agentic AI is the self-driving car: navigating, rerouting, and alerting you if the tire pressure is low. It decides, adapts, and acts.
Rather than waiting to be prompted to âcalculate EBITDA,â an agentic AI monitors monthly performance, identifies anomalies in OPEX, investigates supporting documents, and flags the variance in a report before the review meeting.
Traditional finance tools automate specific tasks (e.g., journal entry matching). Generative AI can draft a risk disclosure note. Agentic AI goes further: it identifies when a forecast is off, investigates the driver (e.g., revenue from a new segment), and does so proactively â without waiting for a prompt.
3-Able to adapt, deconstruct tasks, and act on context. Agentic AI doesn’t wait for detailed instructions. Give it a big-picture goal, and it works out the steps, adapts to change, and responds intelligently to context.
Agentic AI operates with self-direction. It can break down high-level objectives into smaller, actionable tasks. For example, when it is pursuing the goal of âproduct launchâ, it might research competitors, identify key features, draft a launch email, and schedule a social media campaign. The AI agent breaks the goal into these subgoals and handles each accordingly.
Agentic AI can also adapt to setbacks. If a web tool is down or the results arenât as expected, it tries a different method. Crucially, it monitors its environment: spotting changes in context (e.g., a new calendar conflict, budget updates, or updated information) and responds appropriately. It does not blindly follow the original plan.
This isnât a tool that waits for prompts. It works towards a goal, makes decisions along the way, and adjusts course when needed. Itâs more like an intelligent assistant â or even a co-worker.
Ask it to âprepare for the annual budget cycle,â and the agent will break the task down to its parts (contacting budget owners, setting deadlines, identifying departments with late submissions), flag unrealistic assumptions, and update the executive dashboard as numbers change.
So far, three types of agents have emerged in enterprise settings:
Level 1: Task agents. These agents handle specific, well-defined tasks with accuracy and speed. They typically interact with structured data and follow predefined rules. Examples include extracting invoice data from PDFs and uploading it to an ERP system, reconciling transactions for tax purposes, or pulling data from multiple reports into a dashboard. For CFOs, task agents reduce manual work, cut error rates, and deliver efficiency gains at scale.
Level 2: Workflow agents. These agents handle multi-step processes that span departments and systems. They manage handoffs, follow-ups, and exceptions across organizational silos.
Think of an agent coordinating the flow of information between sales, operations, and finance during the budget cycle, triggering alerts when procurement doesnât match purchase orders or delivery schedules, or managing approvals across legal, compliance, and treasury for large capex projects.
Workflow agents promote good governance and meet deadlines without constant oversight.
Level 3: Strategic agents. These agents support higher-order thinking, such as forecasting, simulation, or long-range planning. They integrate data sources, test scenarios, and offer insights, even flagging emerging risks or opportunities.
Examples include running what-if simulations on FX exposure, interest rate shifts, or more exotic tasks such as carbon pricing impacts. They can restructure options based on scenario outcomes or even simulate the effects of M&A deals on capital structure and ROIC. Strategic agents directly enhance decision-making quality and enable proactive strategy.
In short, agentic AI ranges from doing work (task agents) to coordinating work (workflow agents) to thinking about the work and its consequences (strategic agents).
The more autonomous the system, the more trust the company using it must be willing to show in its processes.
While the productivity gains from agentic AI can be significant, success depends on having the right foundations in place, including data readiness, governance, and integration into enterprise workflows.
Bringing the data together
Agentic AI depends on access to structured and unstructured data at scale. In most large organizations, this data is fragmented. Tax-relevant data, for example, is often scattered across ERP systems, spreadsheets, and local databases. Each has its own schema, controls, and level of accessibility.
Without a unified data architecture, AI agents struggle to perform even basic analysis. CFOs can help by facilitating interoperability between systems, encouraging shared data standards, and investing in metadata and access governance. Achieving high-quality data is not a one-off project. It requires ongoing investment in data cleaning, integration, and lifecycle management. CFOs should treat data as a core infrastructure issue, not a technical afterthought.
Putting guardrails around agentic AI
The more autonomous the system, the more trust the company using it must be willing to show in its processes. Agentic AI can deliver efficiency and insight. But, without safeguards, it can also produce flawed recommendations, biased assessments, and opaque decisions.
Finance leaders must ensure that AI agents operate within clear governance frameworks that define the scope of autonomy, set up escalation protocols, and assign human oversight responsibilities. Agents must also be able to show the steps of logic behind their outputs, especially in regulated areas such as financial reporting or tax. Embedding human-in-the-loop models helps balance speed with control. In sensitive domains, agents should act as copilots, offering suggestions. CFOs should work closely with risk, legal, and compliance teams to set ethical and regulatory AI standards.
Before you leap, make sure of a firm landing
The real productivity gains from AI wonât come from headline-grabbing pilots or experimental agents. Theyâll come from sound basics: clean data, clear governance, and close alignment across the business. Without this foundation, even the smartest tools will be of little use.
Agentic AI has the potential to reshape how enterprises operate, not only by reducing cost and increasing efficiency, but by enhancing resilience, transparency, and decision speed across the value chain. For CFOs, this is more than a technology shift. Itâs a chance to lead a broader transformation that creates value for stakeholders and captures value for shareholders.
By deploying agentic AI thoughtfully, CFOs can deliver faster reporting and forecasting for investors, improve compliance and auditability for regulators, streamline operations for internal teams, and enable better, faster service to customers and suppliers. But none of this happens automatically. Real value is unlocked when AI is supported by clean data, clear governance, and business-wide alignment, all areas where the CFO has both visibility and influence. In this new AI-powered landscape, finance leaders need to add a new tool to their arsenal and become architects of intelligent enterprise infrastructure.
Teams using AI are often perceived as less creativeâregardless of their actual output. New research from Cambridge reveals the solution isn't less AI, but better collaboration design. ...
Amid widespread predictions of the bursting of the AI bubble, JosĂŠ Parra Moyano demonstrates how leaders can focus on optimizing initiatives with strong value, data readiness, and employee endorsement. ...
Confusing efficiency with productivity is to mistake speed for direction, and execution for value. Hamilton Mann explains how to avoid the pitfalls in your AI transformation...
Audio available
To create the best customer experience, CXOs should leverage AI to build dynamic systems that adapt in real time to customer needs. We explore how leading companies are using this approach today....
Explore first person business intelligence from top minds curated for a global executive audience