What is an AI agent? Plain-English explanation 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 is the plain-English version.
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 (new email, scheduled time, new record).
- 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, the AI model that reads the input and decides the next action.
- An output, something happens in the real world.
- A log, every action is recorded.
Three examples inside a real firm
1. Invoice filing agent
Trigger: A new email arrives in the AP inbox. Process: The agent reads the PDF, identifies supplier, job vs overhead, the right SharePoint folder. Output: Filed and logged. Human involvement: None, unless confidence is below threshold.
2. Month-end close agent
Trigger: First business day, 7:00 AM. Process: Pulls Xero data, runs reconciliations, flags variances. Output: Exception report to the engagement folder and the senior accountant. Human involvement: 2 hours instead of 2 days.
3. Document chasing agent
Trigger: New engagement opened with outstanding docs. Process: Initial request, follow-ups at day 3/7/14, escalation at day 21. Output: Emails sent, register updated. Human involvement: Only the day-21 escalations.
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 initiating anything.
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.
Josh Jefferd is the founder of Install & Scale, an agency that builds AI agents for accounting, tax and bookkeeping firms. Connect on LinkedIn.