Pigment, which makes planning software for finance teams, offers a free 14-day trial in which you tell an AI agent what you want, such as a budget, a revenue forecast or a cash plan. The agent asks a few questions, then builds a working model and a dashboard you can change and share. No credit card is needed, and after 14 days the workspace becomes read-only. It cannot be bought from the trial; you have to contact the company.
The idea itself is not new at Pigment: paying customers have been able to describe a model in plain words since March 2026. What the trial changes is who can try it, on their own numbers, before anyone signs a contract.
Cheap models change which questions get asked
Think of a regional manager at a distribution business. She asks finance for three scenarios a quarter, such as a fuel price rise, a late-paying customer and a slower month. Each one takes finance about three days to rebuild, so she waits roughly nine working days a quarter and asks for only the scenarios she can justify. If she can describe the scenario herself and get a working model in minutes, the waiting disappears, and she will ask for ten scenarios instead of three, including the odd ones she would never have bothered finance with.
That is what happened with spreadsheets. An accountant in the 1970s could spend a day on one change to a client's numbers, so clients asked for few changes. When a change took seconds, the questions multiplied: what if I hired more people, charged less, borrowed more. The clerks who added up columns lost their jobs, and the accountants who answered the questions became more valuable.
The agent builds the structure, and the assumptions stay yours
The same history carries a warning. Spreadsheets made it easy to build a wrong model quickly: a trader at J.P. Morgan lost $6 billion partly because of a simple error in a formula. An agent that builds a model in minutes will build a wrong one as fast if the description is wrong or the numbers behind it are.
So the skill that counts moves from building to checking. It is the person who knows that the biggest customer always pays late in December, or that the warehouse cannot take more than a certain volume, who can read the assumptions the agent chose and say they are wrong. This is the same pattern we described when we wrote about finance teams getting value from judgment calls rather than cost cuts.
It also says who is exposed. People whose value is configuring models by hand, writing the formulas and wiring the tabs together, are in the clerks' position. People who understand the business well enough to ask the next question are in the accountants' position, and there are likely to be more of them, because every manager who can ask for a scenario becomes a person who needs someone to trust the answer.
If you own a budget, the trial is a cheap way to see this for yourself. Give the agent last year's figures, where you already know what happened, and ask for a scenario whose answer you can predict. If the model gets it right, you can trust it with the question you do not know the answer to. Use only data you are allowed to share outside your company; the trial also offers a sample dataset.