At Google I/O this week, Google announced something that caught my attention — not the flashy consumer stuff, but a quieter announcement about enterprise AI. They’re building agents that run continuously in the background, connected to your internal data, watching for things you told them to watch for. Gemini connected to BigQuery, Google Sheets, Drive, Gmail — all of it — running 24/7 and surfacing anomalies before anyone goes looking.
The example they used was simple: “Alert me when support costs exceed threshold and pull the top categories.” Not a one-time query. A standing instruction. Always running.
That’s a small sentence. But it’s a big shift for accounting and finance.
Why We’ve Always Worked in Cycles
The monthly close, the quarterly report, the annual audit — none of that was a design choice. It was a constraint. If it takes 7 to 14 days to close the books, you can’t do it more frequently than monthly. The whole control framework got built around that reality.
We got rule-based automation along the way. Flag transactions over $X. Catch duplicate invoice numbers. Alert on budget variances. That helped. But those rules are rigid. They catch what you anticipated. They miss what you didn’t.
Think about it like this: a quarterly blood test versus a continuous glucose monitor measures the same thing. But your relationship with the data — and your ability to actually intervene — is completely different. That’s the shift happening right now with AI agents and financial monitoring.
What’s Actually Different About AI Agents
This isn’t just automation with a new name. Rule-based systems catch what you programmed them to catch. AI agents catch what you didn’t anticipate. They adapt as patterns change. They chain across systems — pull from your ERP, match against your bank feed, code to the right GL account, flag the anomaly — all in one workflow.
And here’s something worth being honest about: AI is nondeterministic. That’s a real limitation in finance. You still want deterministic business logic for your core controls. You still want the rules. But the edge cases — the reconciling items that show up at month-end and take hours to track down — that’s exactly where AI can help. It can sit on top of your existing business rules and catch what fell through.
There’s also a data scope point that doesn’t get enough attention. It’s not just structured ERP data. Project details live in Docs. Trackers are in Sheets. Decisions got made in meeting notes. Workspace Intelligence — what Google is building — maps semantic relationships across all of it. That’s a meaningful step toward monitoring the full picture of how financial decisions actually get made.
What Changes for Accountants
Two things, specifically.
Control design shifts from periodic to continuous. If an agent is watching the ledger around the clock, the “we’ll catch it at month-end” assumption breaks down — in a good way and a challenging way. Errors surface faster. But your control framework, your documentation, and your audit evidence all need to account for agent-generated findings. Who reviews the agent’s flags? What happens when a flag goes unactioned? What does the audit trail look like? Those are accounting questions, not IT questions.
Access governance is the new internal controls conversation. The agent’s access scope, what actions it can take, and the human review layer sitting above it — that’s the new control environment. You want to be able to see an audit log showing that 10,000 transactions ran through, 5 anomalies got flagged, and each of those 5 was reviewed and resolved. That’s how you trust and verify a continuous system.
The risk to watch for: firms that bolt agents on top of existing processes without redesigning the oversight layer will create new control gaps while thinking they’re closing old ones. The agent catching an anomaly doesn’t close the loop. A human reviewing and acting on it does.
Key Takeaways
- The shift from “AI you query” to “AI that watches” is the most consequential near-term change to financial monitoring — and it’s arriving in mainstream enterprise tools now.
- Always-on agents don’t eliminate human judgment. They change when and where that judgment gets applied — from finding the anomaly to reviewing what the agent surfaced.
- Your existing control framework was designed for periodic review. If agents are watching continuously, that framework needs to be updated: agent-generated findings, escalation paths, documentation standards.
- Access governance — what the agent can see, what it can do, who reviews its output — is the new internal controls conversation. CPAs should be at that table.
- Firms that design the oversight layer well will run tighter controls with less manual effort. Those that don’t will create new gaps while thinking they’re closing old ones.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply