The Autonomy and Risk Agentic AI Presents to Accounting Firms

Rachel Scanlon and Miklos Orban, Leadership at D2 Legal Technology

AI isn’t just getting smarter; it is now starting to act autonomously. This ‘agentic shift’ is happening right now in accountancy, raising the question: Was Jeremy Hunt correct to say that AI is coming for accountants’ jobs? Rachel Scanlon and Miklos Orban, Leadership at D2 Legal Technology, argue that the risk to accountants isn’t in the use of AI, but rather in its misuse, or even underuse, when compared with other accounting firms that are embracing AI more effectively.

From assistance to autonomy

Accountants already employ AI tools like DARTbot and PairD to assist with tasks such as summarising contracts and audit analysis. While these tools provide valuable efficiencies, they also present several concerns.

The challenges posed by accounting AI tools have led firms to prioritise governance frameworks so that AI-driven work complies with professional and regulatory standards. However, most accountants and auditors still lack the technical knowledge, controls, and oversight mechanisms to deploy AI safely.

Fortunately, a new approach, agentic AI, is emerging to improve technology use within accounting firms. Unlike reactive systems that wait for user input, agentic AI systems can pursue goals independently: reasoning through tasks, adjusting to new information, and carrying out multi-step processes on behalf of users. For firms, this means faster decision-making, more proactive client services and the potential to reimagine entire workflows.

However, with increased autonomy comes a different set of risks. What safeguards will be necessary to ensure agentic AI remains both useful and trustworthy in professional practice?

How Agentic AI works in practice

Consider a private equity transaction. Traditionally, accountants support by drafting a non-disclosure agreement, reviewing financial documents, and advising on the sale and purchase agreement.

With agentic AI, the overarching objective could be to support the transaction team through all stages of the acquisition. The agentic tool would break the transaction down into sub-tasks: preparing due diligence checklists and standardised financial templates, monitoring the data room activity, and flagging anomalies.

Imagine it spots vague disclosures around contingent liabilities. Unlike generative AI, an agentic model will cross-reference past deals, check internal databases and propose adjustments to the purchase price. Simultaneously, it may produce a memo summarising the issue with a risk assessment and recommended stance.

Agentic AI ultimately acts like an autonomous assistant, freeing up the accounting team to focus on high-level deal structuring, negotiations, and client advisory.

Agentic AI: what are the potential risks?

Agentic AI promises efficiency gains, lower costs, and new forms of productivity. Junior team members may see parts of their jobs handed over to intelligent systems, but this will open up the need for new skills and new opportunities in oversight, risk evaluation and client interpretation. This will require a shift in education where future professionals must gain a foundational grasp of data, logic and AI tools.

What’s more, agentic AI uses vast and varied data sources to create a foundation upon which it makes decisions, plans next steps and adapts strategies. This introduces specific challenges, including biases within datasets, misinterpretation of goals, mismanagement of sensitive information and privacy concerns.

When a poor decision is made, who is responsible? What ethical boundaries should guide autonomous action? What are the legal consequences of losses incurred as a result of agentic AI action? These questions are now central to the responsible use of AI and place even more emphasis on firms and users having appropriate training and policy frameworks in place. Implementation must be done cautiously, with safeguards to ensure accuracy and compliance.

Readiness through a governance framework

Agentic AI adoption begins with training. AI fluency must extend across all levels, from senior consultants to trainees, so everyone understands where these systems excel and where their limitations lie. Governance structures and mindsets are equally important to accommodate a new and additional kind of digital colleague. Firms that start small, piloting agentic systems on well-defined accountancy processes, will gain early insights, reduce inefficiencies, and lay the foundation for measurable, sustainable value.