By Piet Baudoin · 27 October 2026
AI becomes part of your office when someone gets room to change how the work is done. Today: stalled pilots, tasks that need no AI, and help with payroll checks.
Why do our AI pilots stay with the same people?
Licenses bought, enthusiastic employees getting started, but the rest of the office carries on as before. Daniel Aardoom of Full Finance, a Dutch accounting consultancy, sees this at accounting firms, he told Dutch trade publication Accountancy Vanmorgen yesterday.
His warning: “De toepassing van AI blijft hangen in experimenteren”. AI adoption remains stuck in experimentation. People in the working group also have their regular billable work to do. He says even a good AI instruction written by one employee does not automatically work for their colleagues.
Aardoom works for a consultancy that helps firms with this. These are his field observations, not a representative study.
What it means for you: a license does not provide time to test a way of working and pass it on. This affects your office too when AI lands on the enthusiast’s shoulders. Also read which processes to automate first.
Source: Accountancy Vanmorgen, 26 October 2026
Do we need an AI agent for every automated task?
HubSpot, a customer management software maker, draws a useful distinction in a guide updated yesterday. Can a task run entirely on predefined data and rules? Then its advice is: “Use a workflow”. In other words, use ordinary automation.
Assigning new inquiries by region can follow fixed rules. Working out which sales opportunity is at risk from call notes requires interpretation. HubSpot suggests an AI agent for that.
What it means for you: the task determines the tool. This is a supplier explaining its own software, not a new product launch. Some capabilities require additional subscriptions or usage credits.
Source: HubSpot, updated 26 October 2026
Can AI flag unusual differences in our payroll?
Oracle announced new AI applications in Fusion, its business software, yesterday. One compares pay, balances, and costs with previous periods. It flags significant differences, with recommended next steps “for review and approval”.
That touches the recurring investigation behind a payroll run. A difference may be correct, but someone needs to find the explanation.
Oracle sells this software itself. This is a product announcement, not an independent measurement of fewer errors or time saved. It does not clarify which Dutch payroll rules this application supports.
Source: Oracle, 26 October 2026
Where are we heading?
Oracle lets employees describe a goal and find the steps needed in their business software (announcement, 26 October). So I expect that within two years the question you start with will matter more than the program you open. You want to know whether you can take on a larger job. Your AI works out what that means for your people, stock, and money, and lays out the options. Within five years, I expect your business software to take shape around decisions like these. Your records automatically supply the background to your plans. You choose where your business goes. What your business already knows becomes an increasing source of new possibilities.
What does this mean for your business?
The common thread is a task that moves the whole team forward. At Bombos, that starts with the shared inbox, for example. Bombos finds the details behind a customer question and drafts a reply. You review and approve it. Our promise: your team starts the morning with an overview of customer questions and replies awaiting approval. That helps the same team do more and better work. That task is the beginning; after that your team teaches Bombos the next one.
Sources
Every source was opened on 27 October 2026 and every quote appears in it word for word.
Accountancy Vanmorgen (26 October 2026) · HubSpot (updated 26 October 2026) · Oracle (26 October 2026)
Bombos team