Field Notes · 09

"Software is dead." What that means for a twelve person business

Mark Cuban says off the shelf software is finished. For a small business the useful question is smaller: which of your tools still fit, and who helps you replace the ones that never did?

Patrick Jean-Baptiste · September 20, 2026

In February 2026, Mark Cuban went on the TBPN podcast and said something that made a lot of software executives uncomfortable. "Software is dead," he said, "because everything's going to be customized to your unique utilization."

It is a headline line, and headline lines deserve suspicion. Your accounting program will open tomorrow morning. Your scheduling system will still send reminders. Nothing died overnight.

Still, there is a real observation inside the exaggeration, and it matters more to a twelve person business than to a twelve thousand person one.

The deal you have been living with

Think about the last piece of business software you bought. Somebody, somewhere, designed it for an imaginary average customer. You are close to that customer in some ways and far from it in others.

So you adapted. You renamed a field because the software had no place for the thing you track. You keep a spreadsheet on the side for the part it cannot handle. Your newest hire spent a week learning the workarounds, which by now feel like part of the job.

That was a fair trade for a long time. Custom software cost tens of thousands of dollars, and renting the average tool for ninety dollars a month was the sensible choice. You bent the business a little to fit the tool, and you got on with your day.

What changed

The expensive part of custom software was always the typing: thousands of lines of routine code, written by hand, by people who bill by the hour. AI tools now write much of that routine code quickly and cheaply.

Two things follow. A small program shaped to one business is now within reach of that business. And software can increasingly read the messy inputs that used to require a person: an email, a PDF work order, a voicemail transcript, a photographed receipt.

Put those together and the old deal reverses. The tool can bend to fit the business.

The part nobody mentions

Here is what the headline skips. Writing code got cheap. Deciding what the code should do is as hard as it ever was, and it is now the whole job.

An AI system knows nothing about your business. It has no idea that your Friday deliveries run on a different route, that one client must always be invoiced on the 15th, or that the "notes" field is where your dispatcher keeps the information that really matters. Feed it a vague instruction and it returns a confident, plausible, wrong answer.

Cuban sees this too. In the same conversation he pointed to the roughly 33 million small and medium sized businesses in the United States that have no AI department and never will. His advice to young people was to learn how AI works, and then learn even more about how to put it to work inside real companies.

That second skill is a kind of translation. Someone has to understand what the technology can do, understand how a specific business runs, and carry meaning between the two without losing anything on the way. Call that person a technical translator. The title matters less than the two fluencies.

Three questions worth an hour of your time

You can start this without hiring anyone.

Where does someone retype something? Any place a person reads information in one system and keys it into another is a candidate. Work orders, intake forms, invoices, appointment requests. List them.

Which subscription do you fight with? Every business has one. Write down the three things you need it to do. If that list is short, a small tool built for you may cost less over three years than the rent.

What should stay human? Decide this on purpose. The call to a family after a hard diagnosis. The judgment about whether to extend credit to a long time customer. Draw that line yourself before a vendor draws it for you.

An owner who walks into any AI conversation with those three lists is already ahead. You know your business in a way no model ever will. That knowledge is the scarce ingredient now.

A caution

None of this is urgent in the way the headlines suggest. A workflow automated badly is worse than one done by hand, because the mistakes arrive faster and look more official. Go one process at a time. Keep a person checking the output until it has earned trust. Ask where your customers' information goes and get the answer in writing.

Software is alive and well. What ended is the assumption that your business has to fit the software. That part is good news, and it is worth a slow, careful look.

Source for the quotes: Benzinga, February 16, 2026.