At 4:45 on a Thursday afternoon, a junior adviser finishes a client memo that once took most of the day. It is clear, structured and confident. The client expects it before five. A senior reviewer has another call waiting. The conclusion sounds right, so the memo leaves the building.
That is where AI changes the pressure inside a professional firm. It can produce a plausible answer before anyone has established whether the answer is sound. The immediate gain is obvious: less drafting time. The harder question is whether the firm has redesigned its review process for that new speed.
A ruling by the District Court of The Hague gives the issue a sharp edge. In ECLI:NL:RBDHA:2026:9527, a legal representative used ChatGPT while preparing a filing that contained incorrect case-law references. The court held that the party using the material remained responsible for checking those references.
This is a business-control issue, not a story about careless professionals or troublesome software. A firm must still stand behind the claims, sources and conclusions it sends to a client, court, tax authority or auditor.
The polished answer is the difficult one
An absurd mistake is often easy to catch. A polished paragraph carrying one wrong assumption is harder. It can survive a quick review, enter a client letter and shape the next decision.
The AFM describes part of this risk as automation bias. Accountants may assess work from an automated tool less critically than identical work prepared by a colleague. The risk grows when the output looks finished, uses familiar language and arrives under deadline.
Technology already sits deep inside audit work. The AFM reports that 91% of regular-licence audit firms used data analysis in statutory audits in 2025, up from 74% in 2022. Generative AI adds another layer. It can draft the explanation, suggest the source and offer the conclusion in one smooth piece of text.
That changes the reviewer’s task. A convincing draft is still a claim that needs testing. Client facts, original sources, legal references, tax assumptions and calculation logic need checking before the work leaves the office.
Return to the Thursday memo. The visible task took less time. The hidden task did not. When review hours disappear from the diary, the firm has not gained efficiency. It has removed part of the service.
Review time belongs in the price
Small professional firms feel this tension sharply. The same partner may win the assignment, supervise the junior, answer the client, approve the invoice and chase payment. AI can relieve drafting pressure, yet it can conceal the moment when review capacity falls below the volume of work being sold.
The AFM has said that time, resources and appropriate fees create room for careful statutory audits. Excessive attention to financial performance can put quality under pressure. The observation reaches beyond audit. Fixed-fee work still needs enough margin for a responsible professional to challenge the output and reconstruct its basis.
Separate production time from assurance time. AI may reduce the first. The second remains part of what the client is buying. A firm that prices work as though both have disappeared may recover the apparent margin later through unbilled corrections, delayed invoices, fee discussions and partner hours.
Work pressure belongs in the same conversation. The Nederlandse Arbeidsinspectie defines it as an imbalance between job demands and a worker’s capacity, including insufficient time to finish work properly. Its 2024-2025 monitor found stress caused by work pressure in 30% of Dutch companies.
For a founder or partner, that makes AI governance a staffing question as well as a technology question. Who has time to challenge the draft? Who can trace the answer back to the original material? When does a deadline turn review into a signature rather than a professional act?
One workflow, two exposures
Accuracy and confidentiality are often handled as separate policy subjects. AI joins them in the same workflow. Before anyone reviews the generated answer, an employee may already have entered a client dispute, payroll issue, tax position or trade secret into an unsuitable tool.
KVK advises businesses to define which AI tools employees may use, for which purposes and with which data. It highlights personal data, trade secrets and sensitive client information. For a small firm, a short boundary that staff can use under pressure is worth more than a long policy that stays unread.
A Rotterdam judgment shows why the data-entry stage matters. In ECLI:NL:RBROT:2026:9319, the court held that, in the circumstances before it, entering potentially privileged information into an external ChatGPT system could amount to disclosure and break its confidential character.
The operational lesson is simple. Map the input before celebrating the output. For a research note, client memo and document summary, the founder should be able to see which tool was used, what information entered it, where the source evidence sits and who accepted the final conclusion.
When that path is unclear, the workflow is unclear.
Responsibility stays with the firm
The EU AI Regulation introduces duties in stages. Transparency provisions concerning chatbots, deepfakes and certain AI interactions took effect on 2 August 2026. The Dutch cabinet has also proposed a supervisory model that relies largely on existing sector supervisors, alongside particular and coordinating roles for the Autoriteit Persoonsgegevens and Rijksinspectie Digitale Infrastructuur.
For most professional firms, the immediate responsibility is more familiar than the regulation debate suggests. Understand the tool. Protect confidential information. Keep the evidence trail intact. Make sure a named professional can stand behind what leaves the office.
Software does not create a new address to which judgment can be forwarded.
Back in the office, the best decision at 4:45 may be to hold the memo until morning. That is not resistance to technology. It is recognition that speed and trust are different products.
AI can help create the first. A responsible professional must still earn the second.
Need a workable review process for AI-assisted client work? We can help set clear checks, roles and time budgets
The data, sourcing, and analysis behind this article were conducted by Paolo Maria Pavan. AI was not used to identify sources, build the factual basis, or produce the analytical judgment contained here. AI was used only as a drafting aid. The final English text was personally reviewed, edited, and approved by Paolo Maria Pavan before publication.
