Two recent stories point in the same direction for firm leaders: as AI agents take on more work in finance, the need for strong oversight grows, not shrinks. At the same time, a tax preparation fraud case is a reminder that client-facing workflows need controls that catch problems early and keep responsibility clear.
Agentic AI can expand capacity, but it also expands the need for oversight
A recent finance industry piece argues that AI agents are multiplying quickly and that many organizations are already struggling to manage them within traditional, manual operating models. For law and accounting firms, that matters because custom AI is no longer just about saving time on isolated tasks; it is about managing a growing digital workforce inside core client workflows.
If your firm is considering agents for research, drafting, intake, or back-office processing, the operating question is not only what the AI can do. It is who reviews its work, how exceptions are handled, and how the firm knows when a workflow has drifted from policy.
Fraud allegations in a tax practice highlight why automation needs controls
The reported fraud case involving a Michigan tax preparation business and two employees is a reminder that client work can go wrong when review and verification are weak. Even where the underlying facts are still being tested in court, the practical lesson for firms is clear: any workflow that touches tax positions, filings, or client representations needs guardrails.
Custom AI can help surface inconsistencies, route unusual items for review, and preserve a record of what was checked. That is especially useful in tax and other high-stakes advisory work, where firms need to reduce errors without losing professional judgment.
Build workflows around review, traceability, and accountable handoffs
For professional-services firms, the strongest AI workflows usually do not try to replace judgment. They support it. That means designing steps where the system gathers information, flags issues, and prepares a draft, while a human owner signs off before anything is sent to a client or filed.
Traceability also matters. Firms should be able to show what inputs were used, what the system produced, and who approved the final output. In practice, that is what makes custom AI usable in a regulated or liability-sensitive environment.
Succession planning and AI adoption face the same management problem
Another recent firm-management story argues that succession planning should not rely on assumptions about who wants to lead. The same point applies to AI rollout. Leaders should not assume a new workflow will be adopted, monitored, or improved just because it is technically available.
Successful automation needs regular conversations, clear ownership, and a feedback loop. If a firm wants custom AI to become part of its operating model, someone must own the process, evaluate outcomes, and update the workflow when client needs or risk levels change.
- Treat custom AI as part of your firm's control environment, not just a productivity tool.
- Use agentic workflows for drafting and triage, but keep human sign-off on high-stakes outputs.
- Build traceability into every workflow so you can review what the system did and why.
- Assign a clear owner for each automation the same way you would for a client service line.
Sources watched
- Michigan Tax Preparation Business Owner Charged with Fraud (CPA Practice Advisor AI)
- Avoid Assumptions: Use Career Conversations to Guide Your Firm's Succession Planning (CPA Practice Advisor AI)
- The New Operating Model in Finance: Managing a Digital Workforce (CPA Practice Advisor AI)
