AI Won't Fix Your Books. Here's What It Actually Does.

Every accountant on LinkedIn is telling you AI will revolutionize your accounting.

I'm going to tell you something different.

I've been in accounting for 28 years. I'm currently building Python automation tools using the Anthropic API. I have one foot in the trenches with clients whose books are a disaster, and one foot in the code, building the tools that are supposed to fix it.

So I can tell you with some authority: AI is not going to fix your books.

Here's what's actually happening.

1. AI can't fix what your systems aren't capturing.

If your field team logs hours in one app, your project manager tracks jobs in a spreadsheet, and your service software doesn't sync to QuickBooks, no amount of AI is going to save you. AI works on data. If the data isn't being captured, isn't structured, or isn't connected to anything, there's nothing for AI to work on.

You can't automate a system that doesn't exist.

2. AI amplifies whatever you already have.

If your bookkeeping is solid, AI will make it faster. If your bookkeeping is a mess, AI will make the mess faster. It does not have judgment. It does not know that the $4,200 your foreman charged to "supplies" was actually a subcontractor payment that should hit COGS on a specific job. It will categorize it the way it's been told to categorize it, and it will do it in 0.3 seconds, and it will be wrong 200 times before anyone notices.

Speed without accuracy is just faster errors.

3. AI is genuinely useful for the boring parts.

Here's where it earns its keep:

• Pulling structured data out of unstructured PDFs (vendor invoices, statements)

• Reconciling transactions where the matching logic is obvious but tedious

• Drafting first-pass categorizations a human can review

• Surfacing anomalies in large datasets that a human would never find scrolling

These are real wins. They save hours. They reduce eye strain. They catch things humans miss.

But notice what they have in common: they all assume your data is already being captured correctly.

4. The owners getting value from AI right now have one thing in common.

Their systems work first.

Their field-to-finance pipeline is connected. Their job costing is set up correctly. Their chart of accounts isn't a graveyard of "Ask My Accountant" entries. They have clean inputs. So when they layer AI on top, AI has something real to work with.

The owners who are frustrated with AI tools — the ones who tried one and got nothing out of it — almost always have a structural problem underneath. The tool isn't the issue. The foundation is.

5. What this means for you.

If you're a $2M–$15M business in trades, construction, HVAC, or manufacturing, here's the order of operations:

First, fix the systems. Get your operational data flowing into your accounting system. Get your job costing right. Get your chart of accounts cleaned up.

Then, layer in automation. Some of it will be AI. A lot of it will just be better-configured software you already own.

Don't skip step one. AI will not save you from skipping step one. It will just make the consequences of skipping it faster and harder to untangle.

I'm building AI tools right now because I see real opportunity in them. But I'm building them on top of clean systems, not as a replacement for them.

If your books feel broken, the answer isn't a smarter robot. It's a real foundation.