Finance teams have been automating for a decade. OCR reads invoices. Rules auto-code the recurring bills. RPA scripts shuttle data between the AP tool and the ledger. And yet the team is still buried at month end, because everything automation couldn't handle became an exception, and exceptions became the job.
That's not a tuning problem. It's a ceiling built into the model.
Rules execute steps. Agents pursue outcomes.
A rule matches a pattern: if vendor is X and amount is under Y, code to account Z. The moment reality deviates (a new vendor, a split invoice, a currency mismatch) the rule stops and a human starts. RPA has the same shape at a larger scale: it replays the clicks a person recorded, and breaks the day the screen changes.
Agentic accounting works from the other end. You give an agent the outcome, reconcile this account, get this invoice posted, and it works the problem the way a staff accountant would: pull the bank feed, compare against the subledger, identify the gap, chase the missing receipt in Slack, draft the correcting entry, and post it if policy allows. The deviation isn't a dead end; it's the work.
Why the exception queue never shrank
The uncomfortable arithmetic of rules-based automation: the easy 70% was never where the time went. Coding a clean, recurring invoice took thirty seconds before automation. The expensive part was always the other 30%: the investigation, the chasing, the judgment-adjacent legwork. Rules removed the cheap work and left the expensive work untouched, which is why a decade of automation barely moved the close date.
Agents attack the 30%. Investigation and follow-up are precisely what they're good at, because those are outcome-shaped tasks, not pattern-shaped ones.
The control question
The usual objection: rules are at least predictable. True, and agents don't abandon that. In agentic accounting, agents operate inside the same approval thresholds, segregation of duties, and audit trail the team already relies on. Every action is logged with what was done and on what evidence. The difference is that the logged actor did the whole job, not the first half of it.
RPA was a bridge: useful, brittle, and always one screen-change from breaking. Agents are what the bridge was pointing at.
Light is the agentic accounting platform. See how agents run finance work end to end.