Workday has announced that its payroll and accounting AI agents will work in the background on their own, watching pay data and financial transactions without waiting for a person to open a chat window. The hiring agent already works this way for early customers, and the payroll and accounting versions are planned for early customers in early 2027.
The deadline stops being the hard part
Picture a payroll specialist at a company of 600 people that has just reorganized three departments. Every backdated pay change has to be checked before the next run, and the check usually means a day or two of opening records to find what is missing. The payroll agent is meant to run that audit itself, flag anything outside the limits the team has set, and draft the notes to the affected employees for the team to review before they go out. For a team that pays twice a month, two days of searching per run is four working days a month that turn into reading a list.
Accounting gets the same change. When a customer signs a contract, the accounting agent can pull out the key details, prepare the entries and propose when the revenue should be counted. It then keeps testing transactions through the year, so the audit becomes a standing check instead of a yearly scramble.
Why checking all the time took until now
The Bell Labs system from 1989 worked on numbers in tables, and a person defined the rules it ran. Anything written in a contract, an email or a note was out of its reach, and most of the work around payroll and the books is written that way. An AI agent reads that text, which is why the idea can finally reach the places where mistakes start.
It helps to be exact about what is new. Workday's accounting agent could already pull details out of documents to prepare billing and entries, and its payroll agent already answered questions about payroll data. Both waited to be asked. What changes is who starts the work.
The new bottleneck is the list of flags
A check that runs every day finds more than one that runs once a month, and each finding needs a person. An agent that flags everything only moves the scramble into an inbox. So the useful skill becomes writing down what normal looks like: how far a retroactive adjustment can differ from the usual before someone looks, which contracts always need a human eye, which gaps in pay data can be fixed without asking anyone.
Workday lets each customer choose, task by task, what the agent does alone, what it prepares for review, and when a person makes the final call, and it logs every action. The hiring agent shows the pattern: it can look for candidates in the background and draft messages to them, but nothing is sent until a recruiter approves it.
The same thing will happen well beyond Workday. Any business that lets an AI watch its invoices, shift pay or stock counts will spend the first week deciding what the agent may do alone. The teams that wrote those rules down before switching it on will get their days back, and the others will get a longer inbox.