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?
Word needs need OpenAI and Anthropic like startups to drive AI forward. Think about only Google, Meta, MS, AWS is only have these capabilities. They will never able to do that in one hand, other hand it will be monopolistics. We need more AI startups, not monopolies.
No wonder why he is saying that, they lost AI game, no top researcher wants to work for IBM. Spent years developing Watson, it is dead. I believe this is a company that should not be existed.
Maybe it's the opposite. IBM spent years on the technology. Watson used neural networks, just not nearly as large. Perhaps they foresaw that it wouldn't scale or that it would plateau.
Sorry forgot to mention B2B AI company. You need be mindful not built everything customers ask you to build. I'm talking about that, email, talking to customers I have no problem with that.
You are on YC and appear to be disagreeing with PG's "Do Things that Don't Scale" article: https://paulgraham.com/ds.html so people jump to conclusions even if you don't.
What is rude? Did I mentioned anything bad here, I expressed my point of view, so other founders can either take it or not. I'm describing a real problem in enterprise AI. If pointing that out makes me "against the cult," so be it. I'd rather share what I learned the hard way than repeat talking points that don't apply to my context.
LayerNext automates your bookkeeping tasks. It automatically categorizes transactions and posts them to QuickBooks when you upload your receipts, bills, or invoices.
At the end of the month, you can upload your bank and credit card statements (PDF) and ask the LayerNext agent to reconcile them.
Once your books are clean, you can also ask financial questions and get insights about your business.
This is neat. Last couple of months, at least 3 people asked me how to select 5 candidate for interview. When we post there are more than 300-500 resumes. If it works as advertised, you will make it.
You can pay this company money to sift through your garbage applications from mainstream job boards or you could just use your brain a little bit to try to attract candidates from places that are biased to above median performers.
In other words, garbage in, garbage out. I think it’s easier to control the input rather than desperately sort through the output. Strategies like employee referrals are known to produce better candidate quality than job board applicants. There are industry conferences, meetups, and other socialization spaces where you’re likely to find higher quality candidates before you even start screening.
Hell, I wouldn’t be surprised a random lottery of the group of candidates with baseline decent resumes isn’t similarly effective, and the best part about that strategy is that it’s free.
I could see this product being good for jobs with low training or experience requirements, but then again once you’re at that level you might as well just go with the random lottery strategy.
Buddhika and Kelum here. We built an AI bookkeeper and CFO because we couldn't afford to close our own books.
Six months ago, we were building financial intelligence for enterprise CFOs. We'd spend months getting access to their SAP or Oracle systems, then more weeks understanding data models that no one documented. After 18 months and a handful of customers, we couldn't make it work reliably. The integration complexity was killing us.
The breaking point came when we realized we hadn't closed our own books for the last full year. Our accountant was chasing us as he needed to file year-end. We use QuickBooks just to invoice customers, never record any expenses or reconcile a bank statement.
So we pivoted and started building a full end-to-end bookkeeping and financial intelligence platform for startups and small businesses. The good thing is at least we can eat our own dogfood and iterate.
Initially we thought it was a super easy problem to solve. Also as engineers we thought this was not complex or sexy enough for us to even work on. But the technical problem was harder than expected. Bank transactions are a mess, "CL GRP INSURED INS" could be what? There are so many edge cases and they're different business to business.
The reconciliation part for expenses was surprisingly straightforward, matching transactions by amount, date and description within a 3-5 day window, flag exceptions. The hard part is handling edge cases like income recognition, refunds, split payments, bulk payments and foreign currency.
What we have now:
- Automatic transaction categorization (QuickBooks integration)
- Email parsing for invoices and receipts
- Real-time cashflow, burn rate P&L that updates as transactions come in
- Month-end close that actually... closes
What we haven't figured out:
- Multi-entity companies (parent/subsidiary structures)
- Handling equity compensation properly
- International tax complexity
- Many more unknow edge cases.
We're looking for 100 founding users who are willing to connect their QuickBooks accounts and try it out. If you're a technical founder doing your own books, I'd genuinely love feedback on what's hardest for you.
Hello hackers,
I'm one of the creators of LayerNext. Our initial product was a metadata lake for images and video for computer vision. The problem was it was difficult to search images or videos from S3 buckets if you have millions of images.
But we realized that computer vision teams don't need a metadata lake. They only need labels :). So we pivoted and added support for all other data types and developed a RAG system to answer any questions from data.
The most difficult part of any RAG system is handling unstructured data like documents. We are laser-focused on handling documents and generating accurate answers.
I appreciate your thoughts. Most of the customers we've talked to have requested a self-hosted version of our platform on their own cloud infrastructure. Regardless of existing agreements like SOC II or Data Processing Agreements, companies are extremely concerned about having their AI data in the hands of third parties. Example, I'm sure that Walmart is not using AWS.
Walmart is one of the largest customers of Azure. They don’t use AWS because Amazon wouldn’t let them and their subsidiaries use AWS back in the earlier days of cloud adoption. Yes, some things are still on prem, but many (and in some cases most) things aren’t.
It can be a bit of marketing. I just want to show the achievement. I bootstrapped this project with a couple of forks. As I said, I wanted to build an Amazon Go like tech with less as I don't have millions. if you are interested how I did it. We used 4 security cameras and fed camera streams to 2 Jetson Orin devices. We developed a 3D perception model to track people and generate events when people pick items from the shelf. No face detection. This is a 8x8 store mock store. hardware cost is less than $5000. The system is ready for deployment in any airport micro store or sports arena. It may be needed to fine-tune the model to fit the environment.
Bottom line, real-time computer vision applications are possible and create value without spending a lot of money.
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?
reply