Most software development was either huge custom business systems or packaged desktop software complex enough to warrant fifty-dollar or more price tags at a time when $50 was worth a lot more than it is today.

Programming was in some ways far more and in other ways far less accessible to get into, and there weren’t software developers around every corner. Someone who could _use_ a computer with a modicum of fluency was a rarity, let alone someone who could program one.

It would be like asking whether accountants should fear for their jobs after the invention of the calculator.

Even MS Access (introduced something like a decade later) or the roughly equivalent FileMaker Pro were ways to build custom software for companies that would never have been able to afford the SWE team to build something from scratch. (And Access in particular tended to be wielded by people unfamiliar with relational database normalization in ways that made it a bit of a time-delayed footgun.)

Access is probably the best analogy for vibe coding in that sense — really good for building something small and custom for someone who couldn’t code it from scratch themselves. Yet not really suited for building large, complex production-grade (let alone “enterprise-grade”) systems — you could try, but you would come to regret it soon enough.

The main difference being that somehow a lot of software companies have managed to fall for the idea that it doesn’t have such limitations.

AI comes at quite a different time, when tech adoption was already pretty mature / saturated to start with.

From the IT side, spreadsheets and later spreadmarts were a completely understandable headache.

The business needed solutions, and IT often couldn’t deliver fast enough. A lot of that was structural: IT was usually treated as a cost center, underfunded, and forced through layers of process and overengineering. So even small things could take too long, cost too much, and come wrapped in too much ceremony.

So the business used spreadsheets because they were powerful, flexible, and already there.

The real problem came later. Business process, business data, and domain knowledge ended up trapped inside undocumented spreadsheets living on someone’s PC. Out of sight, out of mind, and effectively unmaintained.

Then Joe retired, quit, or got laid off, and suddenly some weird but critical business function stopped working because Joe always did it and it lived on Joe’s machine.

That was the nightmare. The same IT group that didn’t have the time or resources to meet the need in the first place now had to reverse engineer a giant kludgy spreadsheet and somehow turn it into something supportable.

Which is part of why the current AI moment feels familiar to me. Not because AI is the same thing as a spreadsheet, but because the adoption pressure comes from the same places: businesses want to save money, people need to get their work done, and they will reach for whatever is powerful, available, and fast enough.

And just like with spreadsheets, some of what gets built this way will be genuinely useful, some of it will become invisible infrastructure, and some of it will turn into a future headache for whoever has to untangle it later.

I don’t think the lesson from spreadsheets was that end-user tools replaced professional software. It was that when the official path is too slow, too expensive, or too disconnected from the real need, people route around it. AI looks to me like the same pattern on a much larger scale. The hype says panacea. Reality will probably be messier.

Early tools like Lotus 1-2-3 and dbase were mind-blowing because they were so generalized and available on consumer appliances. Schools managed milk money, farmers planned crops, the perceived value was instant. There wasn’t an activity that couldn’t benefit.

Back then one computer with a spreadsheet and database was considered more than enough to grow ‘any’ entrepreneurial enterprise from zero to 200 employees. Even in the late 90s I was in a ~300 person multi-national that mostly ran on one Novell server and the entire company lived in Lotus Notes.

There has always been a bit of back and forth. Giving long deadlines and crazy costs, the business will always kruft something together. Sometimes it works, sometimes folks get burned, sometimes you get a nice hand off.

I liken AI to more like sewing machines.

Lets say our factory output was 10 tshirts a day by hand sewing. When sewing machines came around, we din't make 10 tshirts by 8:30am, send everyone home and pat ourselves on the back. Instead we just increased output per worker and kept investing. This drove the cost of goods way down, to the point where a negative pressure equilibrium was reached in the market: making tshirts any cheaper you'd have to give them away (which, 100 years ago was an absolutely absurd proposal, yet today is so common its sort of laughable). This lead to people moving out of the industry but new people moved in (industrial and chemical engineers, material scientists, technicians, mechanics etc).

In my mind, CEOs that are firing their staff are:

1. Making up for past over-hiring mistakes during Covid

2. Cashing future investment in for today's temporary gains.

we're in an economic pinch right now, so not surprised to see layoffs. AI firings are just an excuse; if this happened in 2016, you'd be rolling out AI to every employee in your company to outproduce your competitors. You can't cut your way to success.