Closing books and running payroll feel like solved problems with today’s saas and high stakes if you mess up.

This is one of those areas I would spend more time checking the outputs than it would take me to click the button myself.

What are some of the alternatives that are worthwhile for this?

Gusto makes payroll zero clicks.

I’ve used xero and quickbooks and they integrate with many banks and expense management platforms to automate closing.

Also, unless something changed, unlike other providers, they don't just sell your payroll data to data brokers.

I was thinking about this more and more lately. There is really no escaping this , because even if you are sensible in your choices your vendor or service provider may not be. It will introduce a new level of randomness to our interactions that as a society we may not be quite ready for.

From the more obvious possible issues: no payroll, massive refund overpayment, legally binding agreement that puts the business at disadvantage.

FWIW, I like the idea, but I sure as fuck would not let LLM touch real money or pieces that can allow to move it around.

I suspect that the time spent on accounting or the money spent on accountants will influence decisions made by small-small business owners (1-5 staff range), in that some will take these risks. Admin is a huge pain for very small businesses.

Accountants can be expensive, especially if your books are messy, or have poor accounting practices from the start.

Systems like quickbooks, hubspot, payment processors all have tiers where yes on paper they make it easy to properly setup good accounting practices, but you’ll spend an additional 500/month+ to get those features.

Hiring an accountant to clean up the books and do quarterly book keeping is equally as expensive if not more.

Especially for small service based businesses, where margins can be tight, revenue can fluctuate heavily MoM, committing an additional $6k+ per year just to keep books organized is non-trivial.

As an experiment, i gave all our finance data for 2025 to an agent, and it did quite well after spot checking. There may be a middle ground where users can do exports, verify with “real” software, and have agents handle contextual classification to considerably cut down costs

I dont think they are expensive. I pay 150 eur/month for closing books and payroll with 3 employees. My accountant offered to do the bookkeeping as well for 250 extra. Its a pain to do but not 250 eur pain so I do it myself

Accountants could use AI themselves. Their customers will probably demand lower prices or just ditch them to automate it. It is a bit sad if AI disrupts this field, because it seems like a cooperative strength of humans to organize this synergy.

On the other hand I wonder if it will reveal the downsides of AI at a larger scale. Small businesses will have much lower tolerance for LLM inefficiencies. If it doesn't save time/pain it's just not worth it.

If you think a good accountant is expensive try seeing what a bad one will cost you.

This 1000%. If people want to give this a go then they should be letting QuickBooks attempt to implement this for their customers and they take on the risk. Otherwise this is just asking for Armageddon in the back-office.

As someone in this situation myself who has used AI tools, Claude Code/Codex are useful for doing certain laborious tasks like bookkeeping errors/reconciliation issues but they don't replace a professional accountant.

It's not just about being able to balance Xero but knowing rules, procedures and the way the tax office works.

For how long though? I like my accountant, but I use Claude Code enough to know the SOTA and potential, read through the Claude for Small Business skills and texted a friend "How long do you think accountants have?"

- an Aussie half-wog

Honestly, I think we'll have professional accountants for decades into the future, but they'll become significantly more productive and better at spotting issues.

Claude still isn't at the point where I would personally trust it to be expert level in a field I'm not (very different story when I'm getting it to do something I do know about myself), and the risks of screwing up your reports far outweighs the cost of getting a human to go over things.

But 100%, I can see accountants that use Claude replacing accountants that don't.

(Also, if we're counting, I'm only 1/4 Wog. 3/4 grandparents are Anglos!)

The point of an accountant is accountability. Its in the name. Who do you go after if Claude messes up your books?

>but knowing rules, procedures and the way the tax office works.

Those three letters "CPA" in one's email signature basically expand to "I won't fall for your low effort form letter bluff, you can't get one over on me that easily" as far as the auditor who's following up on the form letter cares.

With a bit of technical knowledge you can get pretty far with accounting without AI or cloud services.

I run a small business (no employees) and GnuCash was ok. Then I got tired of battling it for years to do certain things.

Spent a few days human coding a command line income and expense tracker a little over a year ago at https://github.com/nickjj/plutus.

I do my estimated quarterly taxes with its assistance in literally 5 minutes. All I do is download the CSV files from my bank and run the reports I'm interested in seeing through it. At the end of the year I run through the full numbers and triple check things in about 10-15 minutes. These numbers give me complete confidence to file my taxes accurately from a business income / expense perspective.

Of course you can use the tool for personal income / expense tracking too. Personal vs business is an arbitrary category name.

A lot of "admin" tasks are human problems though. You learn you were supposed to be paid, but it didn't happen, or you were paid the wrong amount.

A computer can help you find that problem, but solving it is still a human issue. One of the things people want to know about invoices is which are likely to be paid on time, as some customers consistently delay or attempt to avoid paying.

> Remember last month when Anthropic was like: no, we're not going to refund you even though we admit we're in the wrong for anti-competitively burning credits?

I'm quite sure at the time that they said they wouldn't give compensation, not that they wouldn't refund them.

For a preview of how this will go, take a look at this:

https://accounting.penrose.com/

Here’s what’s crazy: by making this widely available to SMB, they will soon have enough training data to beat this benchmark —- in probably less than a year is my guess.

True! Between improving training data and figuring out how to provide better context to the LLM, there will be rapid improvement within a year

I tend to agree, but the gains will come at the expense of the early adopters. Then again, this has been the case in so many industries throughout history.

Wish they would update it with the newest models. Probably still fails but would be interesting to see the mistakes getting fewer.

These are solved problems that can be extremely efficiently optimized with ML based solutions that require zero LLMs in the loop. This is business compliance on the line so good luck trusting claude on this. Deranged

LLMs are bad at deterministic output.

Full stop.

Only sane person in this chat.