What Happened to Software is Happening to Finance — AI, Automation, and the Future of Accounting

I’ve never been one to write publicly. I was the type who thought your network was the ability to walk into an office, accomplish something real, and earn relationships as a byproduct. Publishing felt like noise. But something has shifted in the past six weeks that I can’t keep to myself.

What happened to software is now happening to finance.

And if you work in either world, the ground is moving under you.

Background

I’m a CPA. I run finance at a startup. But I’ve also been building software since 1999. It started as a hack — automating the tedious parts of my finance roles. But as those roles became dependent on a stack of disconnected SaaS tools, I realized my finance was only as good as my data. And to get the data right, I had to get good at APIs and data warehouses.

So I became a finance person who codes — a weird niche that turned out to be excellent preparation for what’s happening right now.

The Progression

Here’s the progression I lived on the software side:

A year ago, AI was 20% of my workflow — a better Stack Overflow. I’d use it to research syntax I didn’t understand.

Three months ago, I flipped it to 100% — but I was project-managing every function like if it were a junior developer.

Now I hand it a product spec and let it run. I use Ralph Loops — autonomous iteration cycles where the agent keeps building, testing, and fixing until every requirement passes. I set the vision and review the output. Not the keystrokes.

Projects I used to avoid because I didn’t know a framework? Not scary anymore. The skill isn’t syntax. It’s breaking problems into solvable chunks, managing context, and writing instructions that are reusable. Less typing. More directing. More thinking.

Why I’m Writing This

That same progression is happening in finance.

For a long time, value in finance ops meant being a good human interface to structured systems — Excel, QuickBooks, close checklists, workpapers, the craft rituals (debits left, credits right) — and the ability to wrestle five different platforms into one coherent monthly story. I was so deep in Excel I’ve joked I could compete in the Excel World Championship. (That’s a real thing.)

But “being great at the interface” is losing its edge. Because the interface is becoming an agent.

It’s already showing up in the numbers. Earlier this month, the FT reported that KPMG pressured its own auditor, Grant Thornton, to cut fees — arguing that AI should make the work cheaper. Grant Thornton agreed to a 14% reduction. When a Big Four firm is using AI as leverage to renegotiate its own audit fees, the repricing isn’t theoretical.

I’ve started applying the same patterns that transformed my coding — autonomous loops, specialized agents, structured context — directly to accounting workflows. Inbox triage, invoice validation, reconciliation, exception handling. I wire agents to the things that matter: the accounting system, email, policy documents, and approval chains. And they do it well — if you give them the right context and guardrails. Tools like OpenClaw are making it possible to route work across a fleet of agents, each scoped to a specific job, running on your own infrastructure. The architecture looks more like a well-organized team than a single chatbot.

What Has Leverage Now

Not being the person who clicks, exports, reformats, and reconciles. That’s human middleware — and it’s the first thing agents replace.

The leverage is in designing systems that create auditable truth at startup speed.

Example: for years, a truly auditable monthly close package felt out of reach for a small finance team. Not because I don’t know what “good” looks like — but because when you’re understaffed, you’re constantly trading perfection for survival.

Now I can build a close workflow where the package generates consistently, every number has support attached, exceptions are flagged automatically, and the process is re-runnable next month without heroics. Not a pretty deck. An auditable package. That’s the difference.

Where This Goes

A few convictions:

Data warehouses aren’t dead — but they’re not the finish line. Structured data is still the backbone of accounting. But the missing layer is context: the “why,” the decision history, the policies, the exceptions. The stuff that used to live in someone’s head or a Slack thread. I’ve seen investors describe the next wave as products competing to become a “system of context.” That resonates. The questions that matter are answered in English before they’re answered in SQL: Why did COGS spike? What changed operationally? What’s noise vs. signal?

The future is permissions. The coolest AI demos right now are essentially “dangerously skip permissions.” Fun for prototypes. But in finance, the winners will build checks and balances for agents the same way we build them for humans — approvals, separation of duties, audit trails, accountability.

The full chain matters: inbox to classification to policy to approval to posting to reconciliation to exception handling to audit trail.

The emerging role isn’t “prompt engineer.” It’s the person who can manage an AI workforce — designing workflows, setting controls, maintaining context, and building truth layers so companies move faster without losing governance.

Let’s Talk

If you’re already deep in this: I’d love to compare workflows. There are no best practices yet — we’re all learning by watching each other.

If you’re not in it yet: let’s hang out. I’ll show you what this looks like live. No pitch — just the actual workflows running in real time. DM me.

And yes — I’m being polite to the bots in chat. . . just in case.

Resources that shaped my thinking: