AI cracked the easy part of accounting. The hard part is next.

Every day I go into our San Francisco office, I’m constantly reminded of AI’s tremendous promise. The billboards, the radio ads, the bus wrappers.

Living in the epicenter of the AI boom and working for an AI-first company means I can't escape this question: how much is this actually going to change my job? I get why many are bullish — my team is already using AI to automate complex calculations and generate journal entries automatically. Many CFOs are embracing the change: 54% named integrating AI agents their top finance transformation priority in a recent Deloitte survey. But I also understand why it hasn’t suddenly changed the way we work in the vein of software engineers. Accuracy isn't a nice-to-have; it's the whole job. Tools that get you "close enough" don’t meet our standard.

My take from the controller's seat at Ramp, a tech company doing more than $1 billion in annualized revenue: AI is already shifting what accountants do every day, and it will fundamentally change the profession soon. But it's early days, and the skill and judgment accountants bring will still be required.

Where the progress is real (and genuinely exciting)

Think about a really large, multinational company. They might have 100 people in accounting, but only five to 10 of those people are handling the basics that can largely be automated today: AP, travel and expenses, simple reconciliations, some of the general ledger. Most of the AI accounting features companies are rolling out today are squarely focused on this group.

Accuracy isn't a nice-to-have; it's the whole job. Tools that get you "close enough" don’t meet our standard.

It’s impactful — Ramp’s Accounting Agent, for example, automatically codes and reviews most of our transactions along with many of our accruals. That saves my team dozens of hours every month. But many companies have yet to realize these efficiency gains. Two-thirds of AP teams told PYMNTS their manual work increased from 2024 to 2025, and last year only 14% of North American companies used AI to automate repetitive tasks like invoice matching and reconciliation, according to a Visa report. Still, I expect low-touch to become the standard for this work soon.

These breakthroughs have even bigger implications for smaller and younger companies. If most of your spend flows through invoices and corporate cards, you can automate much of your expense booking all the way through ERP posting. This drastically cuts down on your workload if the volume is manageable and transactions are relatively predictable. This is how even rapidly growing startups can run finance with one person.

But what about the work the other 90% to 95% of the team focuses on at a larger company?

Where AI is only scratching the surface

The day to day looks much different for the rest of the accounting team. They’re doing work that doesn't follow a clean, rules-based process: cash accounting, revenue recognition, GAAP interpretation, equity accounting, and more. Let me dig deeper on a few of these to illustrate why they’re much more difficult to automate.

• Cash accounting: Ramp doesn’t just provide software; we also provide the financial infrastructure to move money. Our platform processes hundreds of millions worth of transactions every day, so we have to map every possible permutation of a transaction into the correct accounting ledger. This equates to hundreds of different scenarios, and these are some of the outliers my team has to painstakingly sort through at month end. We’re actively working on automating this, but it’s not easy.
• GAAP compliance: It’s right there in the name: generally accepted accounting principles. In other words, vague guidelines that leave room for judgment and interpretation. That’s because GAAP applies to all publicly traded companies and most private ones, whether you’re in SaaS, construction, or health care. Two smart accountants can look at the exact same situation involving a revenue recognition question, a complex equity transaction, or a non-standard contract and come to different conclusions. Humans may shift to reviewing AI recommendations, but I don’t see them being cut out of the process.

What I think happens next — and soon

Don’t mistake me for an AI skeptic. I think anything that can be fundamentally captured in rules will eventually be automated, and likely sooner than many expect. It won’t happen overnight, but the pace of change will only pick up from here.

AI will require a different kind of accountant than the one we've traditionally hired. Less expense coding and reporting, more interpretation and decision-making. It’s a shift that should excite accountants.

I'm already hiring with this in mind. I’m evaluating junior hires differently: can this person develop into someone who can own a complex accounting problem and make a call? The skills once critical for leaders will become essential throughout the org chart.

AI will require a different kind of accountant than the one we've traditionally hired. Less expense coding and reporting, more interpretation and decision-making.

Controllers and CAOs need to start thinking this way now. The repetitive tasks that kept those five people busy won't require five people for much longer. But the work that requires genuine expertise isn’t going anywhere, and many people’s jobs are about to get a lot more interesting — and influential.

Melissa Montgomery is VP, Controller at Ramp. She was previously VP, Controller at Patreon and held senior finance roles at Airbnb. Connect with her on LinkedIn.