Agents are good at preparation and chasing, which is where a practice actually loses its hours. They cannot carry liability, and in a profession where someone signs, that boundary is not a preference. It is the business model.
The capability question in accounting is less interesting than the accountability one, and most coverage gets that the wrong way round.
Where the hours actually go
Not the technical work. Practices tend to assume the saving is in preparing the return or the accounts, and some of it is. The larger, less visible loss is in two places.
Waiting. Records that have not arrived. The query sent three weeks ago. Jobs sitting part-complete because one document is missing, occupying mental space and workflow slots.
Chasing. Unbilled, universally disliked, and almost entirely mechanical. A polite reminder on a schedule, escalating appropriately, with a record of what was asked and when.
The second is close to a perfect agent task: repetitive, low judgement, done badly today because nobody enjoys it, and improved simply by being done consistently.
The four that work
| Task | Why it fits |
|---|---|
| Chasing records and queries | Mechanical, unbilled, done inconsistently today |
| Assembling a workpaper from source documents | Verifiable against the source |
| Drafting the query list | A person reviews before it reaches the client |
| Preparing a report or meeting pack | Assembly, not opinion |
All four stop short of the same line: a person reviews before anything reaches a client, and a person signs.
The line that cannot move
Someone carries the liability. A model cannot hold a licence, cannot be insured, and cannot be the person a regulator asks.
That is not a transitional limitation waiting on better technology. It is the structure of a profession where the signature is the product. Any vendor describing an agent as replacing a reviewer is describing something a practice cannot buy, whatever it does.
What genuinely changes is the ratio: less time assembling, the same time reviewing, and some hours released. What a practice does with those hours is the actual decision, and most say advisory and then do not bill it, which is covered in the advisory you are already giving away.
The policy comes before the tooling
Clients are starting to ask what goes into a model and where it is processed, in engagement conversations rather than in audits. A firm with a written answer looks well run. A firm improvising does not.
The realistic failure is not a considered decision to use an unapproved tool. It is a staff member pasting a trial balance into a consumer product at nine at night because it was faster. A policy exists to make that a decision rather than a reflex, and what belongs in one is in what goes in an AI policy for an accounting firm.
The part worth more than the hours
Released capacity only becomes revenue if the advisory work gets scoped, delivered and visible. Most practices find the first two easier than the third: the client cannot see what happened between meetings, so renewal becomes a conversation about price.
GPS for Professional Services is built for that part, with a workspace per client holding their plan, their owners and their progress, so a client can see movement without waiting for a report, and the firm can see every engagement from one screen.
Related reading: AI agents for small businesses, and what AI policy belongs in a practice.
See GPS for accounting firms, and who inherits the client book.
