One thing I keep noticing: compared to programming, accounting often looks like the more automatable problem:
It’s rule-based Double entry, charts of accounts, tax rules, materiality thresholds. For most day-to-day transactions you’re not inventing new logic, you’re applying existing rules.
It’s verifiable The books either balance or they don’t. Ledgers either reconcile or they don’t. There’s almost always a “ground truth” to compare against (bank feeds, statements, prior periods).
It’s boring and repetitive Same vendors, same categories, same patterns every month. Humans hate this work. Software loves it.
With accounting, at least at the small-business level, most of the work feels like:
normalize data from banks / cards / invoices
apply deterministic or configurable rules
surface exceptions for human review
run consistency checks and reports
The truly hard parts (tax strategy, edge cases, messy history, talking to authorities) are a smaller fraction of the total hours but require humans. The grind is in the repetitive, rule-based stuff.
These are all true statements, but all of those things are solvable with classic software. Quickbooks has done this for decades now. The parts of accounting that aren't solvable with classic computing are generally also not solvable by adding LLMs into the mix.
If you want better software, then sure, maybe a coding assistant can help you write it faster, but when it comes to actually doing accounting I would not rely on an LLM in any way shape or form any more than I would do so for law.
The perfect job for AI.
There’s a lot of subjectivity in how GAAP is applied and interpreted - creating accruals, deciding when revenue should be recorded, blah blah.
LLMs can help to handle the subjectivity in how GAAP is applied and provide justifications, which previous rules-based tax software could not before.
That's not good enough when you're dealing with matters that can lead to civil or even criminal liability. Errors can be incredibly expensive to fix, if they can be fixed at all.
With a CPA or attorney, you at least have recourse if they screw up. You don't with LLMs.
We didn't pick this because it was super technical, but because the financial team is the closest team to the CEO which is both overstaffed and overworked at the same time - you have 3-4 days of crunch time for which you retain 6 people to get it done fast.
This was the org which had extremely methodical smart people who constantly told us "We'll buy anything which means I'm not editing spreadsheets during my kids gymnastics class".
The trouble is that the UI that each customer wants has zero overlap with the other, if we actually added a drop-down for each special thing one person wanted, this would look like a cockpit & no new customer would be able to do anything with it.
The AI bit is really making the required interface complexity invisible (but also hard to discover).
In a world where OpenAI is Intel and Anthropic is AMD, we're working on a new Excel.
However, to build something you need to build a high quality message passing co-operating multi-tasking AI kernel & sort of optimize your L1 caches ("context") well.
And it is a very poor fit for moderm LLM based AI. Because accuracy. No mistakes or hallucinations allowed.
I’ve built some software[0] that analyses general ledgers and uses LLMs to call out any compliance issues by looking at transaction and account descriptions.
Is it perfect, nope. But it’s a hell of a lot better than sifting through thousands of transactions manually which accountants do and get wrong all the time.
[0] - https://ledgeroptic.com
I still wonder why humans getting things wrong is a problem, but LLMs getting more things more wrong more often than humans never is. At the very least you'll need a human accountant around to verify the LLM. Or I guess you could just practice "vibe accountancy" and hope things work out but that seems like a worse idea than a trained human professional. But I'm probably just a Luddite.
Also, I am admittedly not an accountant, but I don't think they manually sift through every transaction to verify compliance issues in every single case. That probably isn't how that works.
Our target market is accountants. I want to help them not replace them.
In the same way security audit tools aren’t replacing professionals, but that can help on initial scan.
Some people hate humanity so much that they cannot wait to replace us all with AI so they never have to interact with another human ever again
That's honestly the only reason I can think that they are so biased toward AI
I don’t even want AI to replace us, but it’s a great tool with many use cases.
There might be a way for us to adopt AI as a tool without bringing ruin to many people, but I don't believe that is the goal of anyone building AI.
As it stands, I don't believe there is anything ethical about AI in it's current form. So from that perspective, I vehemently deny there is any value in it
At one point in history, people like you were asking why anyone could be anti-slavery. After all, it was impossible to deny the economic value of slaves.
How can you disagree with the fact?
Some specific examples (like the one you mentioned, _adjacent_ to accounting per se) don't disprove the main point that 100% accuracy is fundamentally impossible with LLMs, while critical for all key accounting aspects.
What I’m working on is the opposite of that. I want to free humans from boring, repetitive finance work so they can use their time for higher-value and more creative things.
While building an “AI CFO” for small businesses (LayerNext), I’ve learned a few things that changed how I see bookkeeping:
Most of bookkeeping is repetitive and under-optimized. Everyone says “90% of the work is repetitive,” but we still hire bookkeepers and bookkeeping firms. Most small businesses I talk to pay around $300–$800 per month just for bookkeeping. Even after paying that, I really doubt every single transaction is recorded in the most tax-optimized way. There are hundreds of transactions, constant government tax rule changes, and limited time.
Current automation is stuck at rules you manually define. Tools like QuickBooks can categorize transactions based on rules you create. That’s it. As soon as something new comes up, you still need a human to either, create a new rule, or manually enter and categorize it.
And even when you hire a human bookkeeper, you still end up doing half the work anyway: sending receipts, answering clarification emails, chasing missing information.
Invoice and expense capture can be 100% automated, even with edge cases In practice, invoice and expense capture is the easy part. With decent models, you can get 100% accurate capture from receipts, PDFs, emails, etc. Edge cases are solvable with better parsing and validation, not more humans.
Reconciliation is the hard part, but reasoning models are getting very good. This is where things get tricky: - multiple invoices paid in a single payment - partial payments - refunds, chargebacks, etc.
For example, imagine a consulting company issuing several invoices to the same customer and receiving one lump-sum payment. We’ve had success using deep research like reasoning to match payments to invoices and handle those cases automatically.
AI can sometimes care more about details than a human. One moment that surprised me.We had a credit card transaction with no receipt.
The question was whether it should be classified as “office expense” or “meals and entertainment” (in Canada these have different tax treatments). When I checked trace of the agent, it looked up the vendor online to understand what they actually sell, checked CRA tax rules and then picked the GL account that maximized the tax benefit for the company.
I’m not sure many manual bookkeepers consistently do that level of research when they’re trying to reconcile 500+ transactions and half the receipts are missing.
My goal is to build a fully automated financial assistant that can close the books without a CPA or bookkeeper, with ~99% accuracy across all transactions, and with the explicit goal of maximizing tax benefits within the rules.
Other outcome is accurate rea-time books can generate good insights to grow the business.
So I don’t see a good reason why small businesses should pay hundreds of dollars per month for humans to do mechanical work that machines can now do, often more consistently and with better attention to tax details.
Curious how others see this, especially CPAs and engineers who have built accounting tools. Is there a fundamental reason we need humans in the loop for the majority of small business bookkeeping, or is it mostly inertia and habit?
That's what LLMs are for. TOOL BUILDERS
And yes , automating that , very valuable.
Really you want an AI interface to a rules engine / system.
Embed with some companies finops teams if you can. Especially the software engineers who are in finops.
If you want complex custom rules, and integration with other systems, you're looking at something like SAP.
can A.I find some edge case deductions. again people out of their depth about certain fields making authoritative statements.
note: I'm passionate about this too. We're building https://fairlight.app/ as a successor to https://moneyflow.camplight.net/ (an internal tool that helps accounting and book-keeping we use to power our venture studio and agency as it's a real hell)
While minor aspects have improved since then, the overall problem indeed remains.
For example, most of what you describe can be solved using Xero in Australia.
Besides, it's from 2016. Unfortunately, I don't own a time machine.
Everything else has been mostly automated since the 90s.
Accounting rules are also not as discrete as programming. There is a lot of discretion. Accounting is basically law with numbers and is corresponding just as difficult/ impossible for LLMs to master.
I've actually be making this comment frequently of late. One of the reasons AI is so popular with c-suite is that most of the time they don't really care if things in the company work right. Just that it makes money.
Accounting will catch any errors that matter. AKA lose money.