Last month I had two engagements land on my desk in the same week that made me rethink what I'm actually paying for in my tech stack.
The first was a year-end for a QBO client. We needed formatted financial statements and a GIFI schedule for the tax software. QuickBooks doesn't produce either in a format that works. Our normal process was to create a ledger in Xero — not because we needed Xero's accounting, but because we needed Xero's reporting engine. We were paying for one platform to format the output of another platform.
The second was an external accountant reviewing a small nonprofit we prepare. They wanted a working trial balance showing the unadjusted balances, the year-end adjustments, and the final adjusted trial balance. The Caseware format. We don't use Caseware. Xero doesn't produce that report. But the accountant needed to see the work in the format they expected — and historically, you'd need a Caseware subscription to produce it.
I built an agent for each. The working trial balance took 10 minutes. The financial statements tool took a couple of hours including corrections. Both now run in minutes. Neither requires a subscription to anything.
You're paying for outputs, not tools
Here's the thing about Caseware, or the Xero workaround, or half the tools in your stack: they do more than formatting — but for many accountants, the formatting is why they're paying. Caseware takes a trial balance and adjusting entries and presents them in a specific layout — columns for unadjusted, adjustments, reclassifications, and adjusted balances. For a lot of firms, that's the job. A valuable one, but a presentation job.
The QBO-to-financial-statements pipeline is the same story. We needed a PDF with an income statement, movements in equity, balance sheet, and notes — branded, formatted, presentation-ready. And we needed a GIFI file — a government-mandated trial balance classification structure — for the tax return. Both are transformations of data we already had.
Once you see the pattern, you start seeing it everywhere. The tool takes your data. It applies a mapping or a template. It produces an output in a format someone expects. You're not paying for the intelligence. You're paying for the presentation layer.
What the agents actually do
The QBO Finalizer reads two Excel exports — the P&L and the balance sheet — and applies a client-specific mapping that translates every QBO account into a financial statement line and a GIFI code. It handles the quirks of QBO's export format — formula strings instead of raw numbers, indent-based account hierarchies, parent accounts with their own balances. It aggregates, formats, and generates both the PDF and the GIFI file directly.
The mapping is a CSV file. Ninety-one rows for this client. Built once, carried forward. Next year, the agent identifies any new accounts that have appeared in QBO, flags them as unmapped, and we add them to the list. The context gets richer over time — client-specific notes on how to handle shareholder loans, which accounts to flag if they have balances, how to present cost of sales for this particular business. That's context engineering in practice. The mapping file isn't just configuration. It's encoded practice knowledge.
The Caseware working trial balance agent does something similar. It takes the closing trial balance from Xero and the year-end journal entries, reverse-engineers the unadjusted balances by backing out the adjustments, and produces a formatted Excel workbook in the exact layout the external accountant expected. Unadjusted column, adjustments column, reclassifications column, adjusted column. The accountant who requested it was satisfied. They didn't ask what tool produced it.
This is what rebuild looks like
Most firms treat their tool subscriptions as fixed costs — line items that get renewed because "that's what we use." The bolt-on approach to AI would be "use AI to make Caseware faster" or "speed up the Xero ledger setup." Preserve the tool, preserve the workflow, shave a few minutes off each step.
The rebuild eliminates the tool. Eliminates the workaround. Builds the output directly from the source data. Fewer steps, fewer dependencies, no recurring cost. A purpose-built agent costs cents per run. The first build takes hours — real hours of mapping accounts, testing output, refining the presentation. Every run after that takes minutes. The investment is front-loaded and the marginal cost approaches zero.
The agent doesn't just replicate what the tool did — it does it in a way that's customized to your practice, your clients, and your presentation standards. And it compounds, because the context files that drive it get richer every year.
Both examples are classic gap workflows — work that happens between platforms, not inside them. The QBO Finalizer works between QBO and the tax software. The working trial balance works between Xero and an external accountant's expectations. No vendor controls these gaps. No vendor can lock you out.
The question you should be asking
Look at your tech stack. Not the core platforms — the add-ons, the utilities, the reporting tools, the format converters. For each one, ask: what does this actually do? If the answer is "it takes data I already have and presents it in a format someone expects" — that's your next agent project.
You're not paying for intelligence. You're paying for presentation. And that's a problem AI solved months ago.