Peter Steinberger @ YC — Hard and Boring

Another idea from Peter Steinberger's recent YC sharing stayed with me.

If he were looking for his next startup opportunity, he said he would pay attention to things that are hard and boring.

The longer we build SnapLedger, the more I understand why.

AI accounting is becoming crowded. Many products can now read an invoice, categorize an expense, generate a report, or put a chatbot in front of accounting data.

Those things are useful.

But they are increasingly not the difficult part.

The difficult part starts when software meets reality.

A customer has a transaction that does not reconcile.

A contract needs revenue to be recognized differently from how it was invoiced.

An expense looks almost identical to another expense, but should receive a completely different accounting treatment.

A VAT rule changes depending on the country, the customer, or the nature of the transaction.

An accountant tells us that what looks technically correct is not how the transaction should really be represented in the books.

These are the problems we increasingly spend our time on.

We work with customers to understand what actually happened in their businesses. When we discover a real problem, we try to fix it quickly — sometimes immediately.

We discuss subtle accounting treatments with our accountant partners. We debate how a rule should work, what the accounting standard actually requires, and how to turn professional judgment into something software can execute reliably next time.

None of this is particularly glamorous.

There is usually no beautiful AI demo at the end.

Sometimes the result of several hours of discussion is simply that one transaction will now be posted to the correct account automatically.

That is hard and boring.

But something important is happening underneath.

Every one of those cases becomes knowledge.

A customer edge case becomes a product rule.

An accountant's judgment becomes system logic.

A reconciliation problem becomes a better workflow.

A compliance requirement becomes something Snappy can eventually handle automatically.

One case by itself looks small.

Thousands of them accumulated over years become very difficult to reproduce.

This is increasingly how I think about the real moat of SnapLedger.

It will probably not be one spectacular AI feature.

It will be the accumulated understanding of thousands of small, messy financial realities — and the ability to turn that understanding into software.

AI makes building software faster.

But it does not remove the need to understand the real world.

And in fields such as accounting, tax and compliance, the real world is full of exceptions, judgment, details and consequences.

That is why I have become increasingly comfortable with the boring work.

Because every boring problem we solve today makes the product slightly harder to copy tomorrow.

Hard and boring compounds.

And over time, that accumulated understanding may become one of the most valuable things we build.

You can watch Peter's YC talk here: Peter Steinberger @ YC.