Andreas Jørgensen is a controller at a company where 95% of the bookkeeping happens without him.
He is also busier and more senior than he was before, because the 95% did not automate itself. He specified it, in plain language, for the way his company's books actually work, and now he reviews what came back and decides what moves next.
"We've automated 95% of bookkeeping with Light and some simple custom agents. This thing is powerful."
Andreas Jørgensen, Controller, Dreamdata
That is the honest version of the answer. The question assumes finance is a fixed body of work that either a person or a machine performs. It never was.
of bookkeeping automated at Dreamdata, specified by the controller who used to do it
people on All Gravy's finance team, which stayed at 3 while the company doubled
average time saved on repetitive admin work at teams running their own Custom Agents
What is actually leaving
Be specific about the work, because the general question produces a general answer and the specific one produces a useful list.
Uploading 500 to 600 card transactions a month by hand, which is what All Gravy's team did before those transactions started landing in the ledger untouched. Keying invoices that arrived by email. Matching a payment to 1 of 4 open invoices 3 weeks after it landed. Downloading balances, assembling a collections list, reading back through correspondence, and writing every chaser individually, which is what Linea at All Gravy describes as having been her afternoon before a dunning agent took it.
Nobody trained for 6 years to do any of that. It was never the job. It was the overhead the job carried because there was no alternative, and it consumed the hours that were supposed to go to the actual work.
What stays, and gets harder
Judgment does not automate, and the volume of it goes up rather than down.
Deciding how a multi-element contract should be recognised. Setting the threshold at which an accrual matters. Determining what an exception actually is, so a system can route it. Working out whether the forecast is wrong or the business is. Explaining to a board why a number moved. Deciding which policies the agents get checked against, and reading the monthly report of what got flagged.
There is also an entirely new category, and it is the one worth paying attention to. Somebody has to specify the automation. That means writing the instruction precisely enough for a system to execute it, knowing the chart of accounts well enough to know when the output is wrong, and owning the result when it runs unattended at 9am on a Monday. That is not a technical skill. It is an accounting skill applied to a new surface, and it cannot be outsourced to engineering because engineering does not know why account 642000 exists.
Leaving
The overhead of the job
Keying, uploading, matching late, chasing by hand, assembling reports from 4 systems. Work nobody was hired for and everybody did.
Staying
Judgment, and more of it
Recognition calls, thresholds, estimates, forecast accuracy, and explaining to the board why a number moved.
New
Specifying the work
Writing the instruction an agent runs on, knowing when its output is wrong, and owning what happens when it runs unattended.
The headcount question, answered with numbers
All Gravy doubled in size while its finance team stayed at 3, and its external bookkeeping partner spent 30 to 40% less time on the books over the same period. Read carefully, that is not a story about people being replaced. Nobody left. The company got twice as big without the finance function getting twice as expensive, which is what growing companies actually need finance automation for.
Ocean.io scaled 2 entities without adding finance headcount. KeyShot's CFO, running $30M+ ARR across 3 entities, put a number on where it goes: "I bet we can get to $100 million in ARR without adding many more resources to it."
The pattern in the data is consistent. Finance teams on this stack are not shrinking. They are staying the same size while the business behind them multiplies, and the composition of their week changes completely.
What the freed time becomes
"We want to remove the time spent on pure transactional work and move it to strategic planning, budgeting, and forecasting. Getting accuracy in the forecast means we can be certain we are spending where we should be."
Sebastian Sandorff Jacobsen, Head of Finance and Ops, All Gravy
At Dreamdata, Peter Egehoved says the agents "allow me time back to do more financial planning and strategy." At Tillo, Harriet Stewart says finance "stopped being a boring back-office function and became a key player in driving growth and direction." At Alva Labs, Thobias Cogrell describes agents doing "the repetitive and manual work for me, and the structured analysis too."
4 companies, 4 phrasings, 1 shift.
The accountants who should be worried
There is a version of this that is genuinely uncomfortable, and pretending otherwise would be dishonest.
An accountant whose value is throughput, processing volume accurately and on time, is holding a skill that is getting cheap quickly. An accountant whose value is judgment, controls design and knowing why the business does what it does is holding a skill that just became more leveraged, because 1 person's judgment can now be applied to the full transaction population instead of a sample of it.
The dividing line is not seniority or age or technical aptitude. It is whether the work depends on the person's hands or their head. Andreas at Dreamdata did not lose his job to the thing that automated 95% of his bookkeeping. He built it.
See the agents that ship with Light, the Custom Agents any admin can write, or book a demo.