“Send Marcos an invoice for $200.”
It's created and delivered via SMS — with a payment link — in seconds.
A contractor texts a photo of a receipt. It turns into a categorized entry. Confirmed in the same thread.
No app required. No login. No accounting knowledge required.
I keep seeing ads and posts — Plaid, fintech thought leaders, industry analysts — asking variations of the same question: “Is agentic accounting the future?”
We stopped asking that question a while ago. We started building for that reality, and we shipped it.
The industry is conflating two very different things
What most people mean by “agentic accounting”: an AI that can answer questions about your financial data. “How much did I spend on supplies?” “What's my revenue this quarter?” This is a query layer. It's useful. It's also trivially replaceable — any LLM with read access to a database can do this tomorrow.
What agentic actually means: the system does things. It doesn't wait for you to ask. It captures, categorizes, invoices, reminds, collects, estimates taxes, and tells you what matters — across whatever channel you're already using.
The gap between these two is enormous.
One is a feature.
The other is an architecture.
Why the input layer matters more than the AI model
We built something called QuickCapture™. It's not a page in our app. It's a universal input layer — any channel you already use, and any channel that's coming. SMS, WhatsApp, Instagram DMs, voice, your web browser. Tomorrow, Siri, Google, Meta Glasses, whatever's next. The system doesn't care how the input arrives.
Why does this matter? Because small business owners don't open accounting software. They never have. And an AI chatbot inside that software doesn't change the behavior. What changes behavior is meeting people where they already are — in their text messages, in their DMs, in a 10-second voice note while they're on a job site.
QuickCapture turns any of those channels into a full accounting action. Not a query. An action.
“Just paid $200 for materials at Home Depot” becomes a categorized expense entry. “Invoice Luis for the fence job, $2,400” becomes a real invoice sent via text with a payment link.
The channel is interchangeable. The capability is the constant.
The architecture has to be different from the ground up
This isn't a chatbot sitting on top of QuickBooks. Bolting an AI agent onto an existing accounting system means you inherit all the assumptions of that system — that the user will log in, that they understand debits and credits, that they'll categorize their own transactions, that “doing your books” is a distinct activity.
An agentic system requires fundamentally different assumptions:
- The user may never log in unless something requires their judgment
- The system should know before being asked — state-first, not prompt-first
- Actions, not answers — “you owe $1,200 in estimated taxes” is information; “want me to set aside $1,200 now?” is agentic
- Multi-channel by default — not a web app with a mobile companion, but a system that treats SMS as a first-class interface
This is where most existing systems hit a wall. They were designed around dashboards and manual input. Adding AI on top doesn't change the foundation. It just makes the dashboard talk.
The real moat isn't the AI
Here's something the “agentic AI” discourse misses entirely: the AI is the least defensible part.
Models get cheaper. APIs get commoditized. If your entire product is “we put an LLM on your financial data,” you have about 18 months before every competitor offers the same thing.
What's actually hard to replicate:
- Data gravity — once your transactions, invoices, receivables, contacts, and tax estimates live in one system, switching costs are real
- Habit loops — a daily text summary you actually read, an action you take from it, reinforced behavior over weeks and months
- Network effects — when your invoice recipient becomes a user, when a lending interaction creates a new account, when every text exchange expands the graph
- Channel infrastructure — a single number that handles accounting, invoicing, and collections across SMS, WhatsApp, and Instagram isn't a feature you toggle on overnight
The companies that win this category won't be the ones with the best AI. They'll be the ones that own the workflows, the data, and the daily habits.
What this shift looks like in practice
Not as a demo. Not as a prototype. This is what it looks like in production today:
- A contractor texts a receipt photo → the system captures it, categorizes it, and either posts it automatically or queues it for quick confirmation — depending on confidence
- A business owner asks “how did I do this month?” via SMS → gets a plain-language P&L summary with actionable next steps
- “Send an invoice to Maria for $400” via text → invoice created, delivered to Maria via SMS, Maria can pay or respond in the same thread
- Tax estimates calculated proactively, with nudges to set money aside — before the owner thinks to ask
- All of it works in English and Spanish, because that's who the users are
No one had to open an app. No one had to learn accounting. The system just ran in the background.
The category is being created right now
When fintech companies run ads asking “is agentic accounting what's coming?” — they're spending money to educate the market on a shift that's already underway.
In two to three years, every accounting tool will claim to be “agentic.” The ones that will actually deliver on that promise are the ones that were architecturally built for it from day one — not the ones that added a chatbot to a ledger.
This isn't a prediction. It's a pattern we've seen play out in every software category AI touches: incumbents add a layer, native builders rebuild the stack.
The future of small business accounting isn't software you use. It's a system that runs — and you check in when it matters.
MarathonBooks™ is built around this idea — your books run automatically, and you check in when it matters.
If you're curious what this looks like in practice: marathonbooks.app