The Expert Said Don't Learn To Code

In 1997 I sat down in front of a beige box running Windows 95, opened QBasic, and started learning to program.

The machine was not connected to anything. There was no web in our house. There were no smartphones anywhere, because there was no such thing. If I wanted to know how to do something, I had a help file and whatever books I could get my hands on.

The kid at that keyboard had no concept of the job he would end up doing. Not a vague idea he hadn’t fleshed out yet. No concept at all.

Before That

Computers, to me, started as a Commodore 64 and then DOS.

You turned the machine on and it was yours. Whatever was on it was on it. Whatever wasn’t, wasn’t, and no amount of wanting changed that. Software arrived on physical media, carried by a human being, and if it was broken it stayed broken.

Then Windows 95 landed and everything looked different, but the fundamental situation hadn’t changed at all. The computer was still an island. It did what it did, on its own, in a room.

That was normal. That was just what a computer was.

Getting Online

Around 1998 or 1999 we got dialup, sort of.

We didn’t dial an ISP. We dialled into my dad’s work, and his work then routed us out to the internet. Technically it worked. Practically it was a series of compromises stacked on top of each other.

Our speed was capped by the modem at the other end, not ours, and whatever bandwidth the office had going out was shared with everyone already using it. So you were never getting the speed on the box. You were getting whatever was left over after an entire workplace had taken its share.

And there were only two or three modems in the pool. If staff were on them, you weren’t getting in at all. You’d dial, get a busy signal, wait, and try again. Getting online was not a thing you did, it was a thing you attempted.

Paying By The Hour

Eventually we got our own connection, which felt like an enormous upgrade right up until the first bill.

You paid by the hour. Not a plan, not a data cap, an hourly rate for the privilege of being connected. A 33k modem, billed like a taxi meter.

You learned to be efficient in a way that has completely disappeared. You knew what you were going online to do before you dialled. You found it, you got it, you disconnected. Idle browsing was money.

We upgraded to 56k, still paying by the hour, and ran it on the free AOL trial CDs that arrived in the mail constantly. There were so many of them. They came with magazines, they came with nothing, they just appeared. Free hours, over and over, from a company we were never going to become customers of.

The Ladder

ADSL arrived for us around 2002 at 256k, and it was genuinely transformative, mostly because it was always on. No dialling. No busy tone. No meter running.

After that it was just a ladder. 512k. 1.5Mbps. 8Mbps. 24Mbps. Then the NBN, starting at 12Mbps and climbing through 25, 50, 100, and now 1000.

In 1999 I was sharing a fraction of an office modem. In 2025 I have a gigabit into the house. Same person, same desk, one working life.

Nobody sat me down at any point and explained that this was going to happen. It happened one incremental upgrade at a time, each one feeling like the sensible next step, and only when you line them up does the size of it land.

What I Built While The Pipes Grew

The work moved with it, and I didn’t notice that either.

I started writing DOS programs. Then Windows programs. Then websites, which at the beginning were barely programs at all. Then web applications that were absolutely programs, and did things I would have called impossible a few years earlier.

Then mobile apps.

That’s the one that still gets me. Go back and tell the kid in 1997 that he will spend part of his career writing software that runs on a telephone. That isn’t an ambitious prediction, it’s nonsense. Phones were bolted to walls. Phones were for phone calls. The sentence has no meaning.

Not “unlikely”. Not “a long way off”. Meaningless.

The Expert

In 2001, an industry expert came to my school to talk to us about our futures.

The timing matters. The dot-com bubble had burst the year before. Companies that had been printing money in 1999 were gone, tech stocks had collapsed, and the mood everywhere was that the whole thing had been smoke. So he wasn’t speaking into a vacuum. He was walking into a room of teenagers a year after the most visible tech failure any of us had lived through, and he had the wreckage to point at.

His message was blunt. Anyone would be stupid to get into programming. Programmers are dime a dozen, there are too many of them already, and they will be out of work. If you want a career, get into servers and infrastructure. That work has to be done on site. You can’t move infrastructure away from the business.

He was confident. He had the room, and he had a fresh crash backing him up. To be fair to him, the logic was reasonable if you accepted the premise, which most of that room did.

I didn’t. I just doubted him.

Twenty-something years later, the infrastructure is the thing that left.

It left the building, left the city, and in a lot of cases left the country. It became a line item on a card. The one asset he identified as permanently anchored to the office turned out to be the single most portable thing in the entire industry.

And the programming, the thing that was supposedly commoditised into worthlessness, is still here. Still in demand. Still paying.

He didn’t just get it wrong. He picked precisely the wrong horse, for a reason that sounded airtight at the time.

I don’t tell that story because I was clever. I was a teenager with a hunch. I tell it because he was the one with the credentials, the slides and the confidence, and the room believed him, and he was comprehensively wrong about the one thing he’d been brought in to be right about.

And Then It Kept Happening

That wasn’t a one-off, it was just the first one I sat through in person.

The crash he’d built his case on didn’t mean what everyone said it meant either. Get out of tech, pick something stable, people genuinely said this to me while I was starting out. What actually died was a pile of companies with no revenue, and the internet they were all built on carried on growing without them. By the time I was properly working, the shortage was worse than before it burst.

Then it was offshoring. Every publication ran the same piece: your job is going overseas, the rates aren’t close, the maths is undeniable, retrain for something that can’t be shipped out. Some work did go. A lot of it came back, because the hard part had never been the typing.

Then low-code and no-code, and citizen developers who would build their own applications with drag and drop. They did build things. Then those things hit a wall, or broke, or needed to talk to something else, and a developer got called.

Every one of those arrived certain. Every one of them turned out to be a tool that changed what the work looked like without removing the need for someone who understood it. Each abstraction moved the work up a level rather than deleting it.

So What About AI

Now it’s AI, and I use it every day. It’s genuinely useful, it’s improving fast, and I’m not going to pretend otherwise.

The forecasts have started again, and they have the same shape as all the others. Confident, specific, and built on the assumption that whatever feels solid right now will stay solid.

That’s the part worth paying attention to. The man in 2001 wasn’t wrong because he was stupid. He was wrong because he looked at the most physical, permanent-feeling thing in the industry, racks of hardware bolted into a room, and assumed it was the fixed point. It was the only thing that moved.

Every prediction I’ve been handed in thirty years has really been a description of the present with the numbers turned up. That’s the whole trick, and it’s why they keep missing.

The future doesn’t arrive as more of today. It arrives as something that didn’t have a name yet. Nobody predicted mobile apps by extrapolating from faster modems, because a phone you write software for isn’t a faster modem, it’s a different category. The change that mattered was never on the axis anyone was measuring.

So I don’t know what AI turns into, and neither does anyone selling you a picture of it. But I know what to do in the meantime, because it’s worked every time so far.

What Actually Held Up

Every one of these shifts moved the work up a level rather than deleting it. Assembly to C, hand-written HTML to frameworks, servers in a cupboard to somebody else’s data centre. The typing got easier and the job got bigger.

What survived each move wasn’t a technology. It was the ability to work out what’s actually needed, decide what not to build, and be the person who understands the thing when it breaks. None of that has ever been automated away, and all of it got more valuable each time the tools got better.

That’s not a hedge. It’s the one pattern with a thirty-year track record behind it, and it’s why I’m not spending much energy on the current round of obituaries.

From A Busy Signal To A Gigabit

I’ve been at a keyboard for nearly thirty years. My connection to the world went from fighting staff for one of three office modems to a gigabit into the house. My programs went from a DOS prompt to a device in my pocket that didn’t exist as a concept when I started.

None of that was on anyone’s roadmap. It happened anyway, and the people who did well out of it weren’t the ones with the best forecast. They were the ones still building when it showed up.

So when someone tells me they know exactly what development looks like in 2046, I think of a very confident man in 2001, telling a room full of kids to go and specialise in the one thing that was about to leave the building.

I’ll take my chances with the code.