Think about the last folder of client documents you received. Four dozen scans, receipts, and forms with names such as “IMG_4471.pdf,” covering a personal return, a side business, and a rental property. Sorting it out might take a staff member an hour. An AI agent can do it in a few minutes.
That is the shift accountants need to understand right now. AI is no longer just a chat window where you type a question and get an answer. It is becoming a coworker that can take on real, multistep tasks.
When ChatGPT arrived in late 2022, AI meant a conversation: one prompt, one reply. Since 2025, the bigger story has been agents. An agent takes a set of instructions, breaks a task into steps and carries them out, often by working directly inside other applications. It can read an inbox, look up a record in a customer relationship management (CRM) system, pull a report from QuickBooks, or reorganize a folder of files.
Measured against the adoption curve of the internet, agents are roughly where the web was in 1993 or 1994. If you are experimenting now, you are early, not late.
A skill is a text file of instructions that tells an agent how to do a specific job. Think of it as a standard operating procedure for a digital employee: how your firm names client files, how it prices a tax engagement, what steps it follows to reconcile an account. Write it once and run it as often as you need. A good rule of thumb: If you have done the same task with an AI tool twice, turn it into a skill.
A connector gives an agent access to the applications where your data lives, such as email, document storage, the general ledger, or practice management software. The real power comes from combining them. Ask a connected agent about the status of a prospect, and it can find the email thread, notice the person is missing from your CRM, and create the contact record from the email signature, all in one pass.
You don’t need to be technical to write good instructions. It is a lot like delegating to a new hire:
Client data raises the first question for most firms. Treat AI platforms the way you already treat cloud accounting or document storage software: as vendors trusted with sensitive data under contract. Here are a few essentials:
The professional lines haven’t moved. AI output is your firm’s work product, and the liability is yours. A human should review anything that touches the books, a return, or a financial statement. Don’t assume your professional liability policy covers AI errors; ask your carrier. And whether using AI in tax preparation triggers the Internal Revenue Code Section 7216 consent requirement is still a gray area, so get qualified advice and make a deliberate decision.
The biggest risk isn’t picking the wrong tool. It is paralysis. Start small and low-stakes:
And skip the question “How do we use more AI?” Ask instead: Where is our time being wasted, where are clients waiting on us, and what are we still doing by hand?
For bigger decisions, use a simple test. Could one well-written skill do this job? Does a vendor you already use offer it? What is the “blast radius” if it goes wrong? High-stakes work such as core tax preparation or audit testing belongs with established platforms. Lower-stakes work around the edges is where firms can build their own.
AI can handle a growing share of routine work, but it can’t replace professional instinct: the sense that numbers don’t add up, or that a defensible position is wrong for a particular client. Protect it. Think a problem through before turning to AI, treat the tool as a sparring partner rather than an oracle, and decide deliberately what you will never automate.
Clients don’t ultimately buy bookkeeping or tax returns. They buy security, clarity about their finances, and help building a better business. The tools will keep changing. The accountants who learn them now, use them responsibly, and keep their judgment at the center will be the ones clients rely on most.
Toni Witt is the co-founder and CEO of Sweet. He is the fractional AI director for dozens of accounting firms and builds AI systems that help firms win real hours back. He has been building AI tools for accountants for over three years and has trained over 2,000 accountants in AI skills.
This article is provided as a complimentary resource by the PICPA and Toni Witt. Statements of fact and opinion are the author’s responsibility alone and do not imply an opinion on the part of CPA Crossings officers or members. The information contained herein does not constitute accounting, legal, or professional advice. For actionable advice, you must engage or consult with a qualified professional.