A CPA firm in 2026 is doing more work with fewer people, against the same immovable deadlines. The pipeline of new accountants is thinning, busy season still ends when it always did, and the software you bought to make it easier mostly moved the work around rather than taking it off anyone’s desk. So when someone tells you AI will fix this, the right first reaction is skepticism. You have heard that promise before, about other tools, and you are the one who paid for the gap between the pitch and the season.
The useful question is not whether to “adopt AI.” It is narrower and more practical: what can AI reliably prepare for your team, and what must a person always decide? Get that line right and the rest of the conversation becomes manageable. Get it wrong and you have bought another experiment that creates more work for the staff you already cannot spare.
Before the answer, our interest stated plainly: Gridex operates this kind of managed AI capacity for accounting firms, so we are an interested party in the argument that follows. That is exactly why every claim in it is named, sourced, and checkable.
The Firm Has Fewer People and the Same Deadlines
Start with the constraint, because it is the whole reason this question matters. The supply of new accountants has been shrinking for years: AICPA data reported through the Journal of Accountancy puts accounting degree completions at roughly a multi-decade low, and the number of new candidates sitting for the CPA exam has fallen to about 28,000 in 2024, down from a high of roughly 48,000 new candidates in 2016. The Bureau of Labor Statistics still projects roughly 124,000 openings a year for accountants and auditors through 2034 — most of them replacing people who leave. The deadlines, meanwhile, do not move. April 15 is April 15.
That gap is where the day actually goes. Not into the licensed work — the returns, the audits, the advisory — but into everything upstream of it: collecting documents that arrive late and scattered, sorting them, figuring out what is missing, chasing it, and deciding whether a file is finally ready for a preparer to open. The document chase is its own problem, and it is worth understanding on its own; the short version is that it is judgment-light, attention-heavy, and it lands on exactly the staff you most need for higher-value work.
The Useful Question Isn’t “Should We Use AI”
Here is the reframe. Most of the disappointment with software in firms comes from a category error: treating a tool as if it could carry judgment. It cannot, and it should not. What it can do is prepare — take a pile of inputs and turn it into something a person can act on quickly.
So draw the line explicitly. On one side is preparation: reading, sorting, extracting, drafting, tracking, flagging. On the other side is the decision: the tax treatment, the materiality call, the professional skepticism, the signature. AI belongs entirely on the first side. The moment work crosses to the second side, a credentialed person owns it — not as a courtesy, but as a requirement.
What AI Can Prepare — and What It Must Not Decide
| Work in a CPA firm | AI can prepare | A person must decide |
|---|---|---|
| Client document intake | Sort what arrived, flag what’s missing, draft the follow-up | Whether the file is truly ready for a preparer |
| Source-document review | Extract figures, tie them to a checklist, cite the source page | The tax or accounting treatment |
| Missing-item follow-up | Draft the chase message, track who’s outstanding | Tone with a sensitive client; when to escalate to a partner |
| Workpaper prep | Assemble a review-ready brief with exceptions flagged | Sign-off, professional skepticism, the final position |
| Recurring admin queues | Run the queue, surface the exceptions | Every exception, and the client relationship |
Read the table by its second and third columns. Everything in the middle is real work your staff does today by hand, and none of it is the work your firm is licensed for. Everything on the right is the work only your people can do — and the point of preparing the middle column well is to give them more room for the right one. CPA.com’s 2025 report on AI in accounting lands in the same place: the technology changes what an accountant does day to day without removing the requirement that a person stays in the loop.
Packet, Queue, Brief: Three Words Worth Knowing
You will hear these three words a lot, so it is worth pinning down what they mean in plain terms. None of them is software you log into. Each is a prepared thing a reviewer opens.
- A packet is a single review-ready bundle for one file or one client. A client document intake packet, for example, shows what arrived, what is still missing, the facts pulled from each document with a link back to the source page, a drafted follow-up ready to send, and anything flagged for a human. The reviewer opens one prepared thing instead of a portal, an inbox, and a pile of paper.
- A queue is a running list of the same recurring task across many clients, kept current and sorted by what needs attention. A missing-item queue across every open return shows who is outstanding and for how long — so nobody re-checks portals one client at a time.
- A brief is a short, structured summary that turns a stack of inputs into something a person can act on in minutes. A review-ready workpaper brief ties figures to a checklist, calls out exceptions, and cites sources — so the reviewer spends time on judgment, not assembly.
The throughline is simple: a packet prepares one file, a queue tracks many of the same task, a brief compresses inputs into a decision-ready summary. None of them makes the decision.
Where the Human Line Sits
This is the part that has to be airtight, because it is where firms get burned. The professional standards already answer the boundary question, and they answer it in your favor.
The AICPA’s Statements on Standards for Tax Services, revised effective January 2024, now name AI explicitly as a tool — and they keep the responsibility on the practitioner who signs. Using a tool does not transfer the duty to apply professional judgment, and it does not transfer the liability. IRS Circular 230 §10.22 points the same direction: a practitioner must exercise due diligence as to accuracy, and may rely on another’s work only with reasonable care in supervising and evaluating it. AI output is exactly the kind of work that standard governs. In plain terms: whatever prepared the file, the person who signs still owns the judgment, the skepticism, and the consequences.
AI can prepare the file. It cannot decide the position, and it cannot carry the liability. The signing practitioner does — every time, regardless of what tools touched the work along the way. Good systems are built to that line, not across it.
What This Looks Like When Someone Operates It
So far this has been about the work, not a vendor. That is deliberate, because the work is the point. But it is fair to ask what it looks like when someone runs this for you rather than handing you yet another login.
For this kind of work, Gridex usually combines three capabilities. Managed Intake turns whatever arrives — portal, email, paper — into a structured, qualified file. Document Review & Research turns those documents into a source-cited, review-ready brief. And workflow automation runs the recurring missing-item and follow-up queues so they stay current without burning a staffer’s week. Underneath all of it sits human review and an audit trail of what came in and when. None of this asks your firm to adopt a new platform or relearn how the engagement is run; a managed team absorbs the upstream sorting, chasing, and reconciliation, and hands back something finished. What reaches your desk is a packet or a queue, not software to administer — and the judgment line stays exactly where the standards put it. Your reviewers still make every professional call.
Where to Start
You do not need to overhaul a firm to answer the question this article started with. Pick the one recurring block that hurts most. For most firms that is document intake — it is the first dam in the season, and the one a finished intake packet most visibly clears. Prepare that well, keep every decision with the people licensed to make it, and you will have learned more about what AI can actually do for your firm than any demo will teach you.
Gridex turns work demand into AI capacity. For accounting firms, that means preparing the upstream of the engagement — intake, review, and the recurring follow-up queues — and handing your reviewers something ready, so staff spend less time sorting and more time on the judgment only they can provide. See how it fits CPA firms or tell us where your team is buried.