AI for accounting firms, explained the way a partner actually needs it.
Everyone is selling you AI. Almost no one will tell you which roles it can quietly take over inside your firm this quarter, what it will cost, and what it won't touch. This is that resource, written for owners, not for tech buyers.
A whole library, written for firm owners.
Every post answers a question a partner actually types into Google. No vendor pitches, no consultancy buzzwords, no em dashes.
The honest answer, what gets automated first, and what stays human.
How to evaluate Digits, Basis, ChatGPT, and purpose-built agents for your firm.
What the top tax-focused AI tools actually do, and which firms they're built for.
A 7-question scorecard partners use before signing anything.
Where to start, what to pilot, and the guardrails to put in first.
Categorisation, reconciliation, chasing, what AI does well today, and where it still fails.
Pricing, packaging and AI-powered delivery, the 4 levers that actually move firms.
AP, month-end close and document chasing, ranked by ROI and risk.
Data access, audit trail, accountability, change management, and trust.
Browse every article in the library.
What is an AI agent? Plain English, for accounting firms.
Everyone in the accounting sector is talking about AI agents. Most people using the term cannot explain exactly what one is, or what makes it different from the AI tools they have already tried.
This post is the plain-English version. What an AI agent actually is, what makes it different from ChatGPT, and what it looks like running inside a real accounting firm.
The definition
An AI agent is a system that takes a goal, breaks it into steps, uses tools to execute those steps, and completes a task without a human in the loop for every action.
The key word is autonomous. Not "answers a question when asked." Autonomous, it runs based on triggers, not prompts.
Invoice arrives in the AP inbox
Agent reads, classifies, assigns job code
Filed to SharePoint, action logged
Hover or scroll to advance the diagram. This is what makes an agent different from ChatGPT, it runs on triggers, not prompts.
What makes it different from ChatGPT
ChatGPT, Claude in a browser tab, any LLM chat interface, these are reactive. They respond when you ask. They stop when you stop. They have no connection to your systems, no ability to take actions in the world, and no memory between sessions unless you provide it.
An agent is proactive. It has:
- A trigger, an event that sets it running, a new email, a scheduled time, a new record in a system.
- Tools, connections to real software, Outlook, SharePoint, Xero, MYOB, Salesforce.
- A goal, defined in plain language: "classify this invoice and file it to the correct folder."
- A reasoning layer, an AI model that reads the input, makes a decision, and determines the correct action.
- An output, something happens in the real world, a file is moved, an email is sent, a record is updated.
- A log, every action is recorded.
When you combine these components, you get a system that does work without being asked. That is the fundamental difference.
What this looks like in an accounting firm, three examples
Invoice filing agent
- Trigger
- A new email arrives in the accounts payable inbox.
- Process
- The agent reads the attached PDF, identifies the supplier, determines whether the invoice is job-related or overhead, identifies the correct job code if applicable, and determines the correct SharePoint folder.
- Output
- The invoice is filed to the correct folder with the correct naming convention. A log entry is created.
- Human involvement
- None, unless the agent's confidence is below a defined threshold, in which case it flags for review.
Month-end close agent
- Trigger
- First business day of the month, 7:00 AM.
- Process
- The agent connects to Xero, pulls transaction data for the closed period, runs reconciliation checks against each balance sheet account, compares current period to prior period, identifies variances above the defined threshold.
- Output
- A structured exception report is placed in the engagement team's SharePoint folder and emailed to the senior accountant.
- Human involvement
- Review of the exception report. Approximately 2 hours instead of 2 days.
Document chasing agent
- Trigger
- A new engagement is opened in the practice management system with outstanding documents listed.
- Process
- The agent sends the initial document request. At day 3, 7, and 14, it sends personalised follow-ups. At day 21, it escalates to the engagement manager.
- Output
- Emails sent, responses logged, register updated when documents arrive.
- Human involvement
- Only for day-21 escalations, the cases that actually require a conversation.
The one-sentence summary
An AI agent does the work you would otherwise have to prompt an AI to do, but it does it automatically, connected to your real systems, without you having to initiate anything.
That is what separates agents from chat tools, workflow automation, and any software you buy off the shelf.
Get a Free AI ROI Audit for your firm
See exactly which roles in your accounting firm AI agents can replace, and how much payroll you'd recover in the first 90 days.
No call required to get the audit.
Why this matters for your firm right now
This isn't a technology story. It's a margin story.
Firms that automate the back office in 2026 will quietly out-price you in 2027. The buyer doesn't see the agent, only the lower fee.
The graduate pipeline is shrinking. Agents close the gap without you having to win a recruiting war you can't fund.
Your SME clients have ChatGPT on their phones. They're already asking why their invoices still take 5 days to file.
Not panic. The quiet urgency of someone watching a flight board change to 'last call.' There's still time. Not infinite time.
Josh Jefferd is the founder of Install & Scale, an agency that builds AI agents for accounting and professional services firms across Australia. Based in Melbourne.
Connect on LinkedIn.