The Day We Forgot How to Code

Photo by Gwendal Bar on Unsplash

I didn’t think that it would happen so soon. I didn’t think it would happen today. Yet because of a Claude outage during peak working hours, our dependency on AI (and worse a particular flavor of AI) has been exposed.

Yes. We could have simply fired up Cursor and everything would have been fine. However, we decided to raise an incident, and that tells its own story, and it’s not good (is it).

A weird disaster

Another day at the office. Another incident. They seem to be more frequent these days, but it’s unclear whether we should lay the blame at the door of AI because ChatGPT tells me that correlation does not equal causation.

The engineering team went into meltdown. The incident was raised and the “important” people sat in a room.

Within minutes people were discussing alternatives.

“Just use Codex”

That was the first suggestion.

Someone pointed out that maybe incidents would take longer to resolve while Claude was unavailable.

Finally somebody made the joke that landed a little too close to home.

“I thought engineers wouldn’t have forgotten how to code”

Everyone laughed.

But only because there was more truth in it than anyone wanted to admit.

A Race to the Bottom

If you’d told me two years ago that an AI outage could slow down a software engineering team, I’d have laughed too.

Developers have survived outages of package repositories, CI systems, cloud providers, documentation websites and Stack Overflow. We always found a way.

Today feels different.

AI isn’t replacing software engineers. It’s replacing our instinct to think. Our ideas. Our positive behaviors. The syntax recall. Any will to learn regex. All gone.

Replaced with context switching. Five minutes here. Ten minutes there. Twenty browser tabs. A trip through old pull requests. Half an hour searching documentation because you know you’ve solved this before.

Claude, Codex and ChatGPT have quietly become the developer sitting next to us.

Not because they’re always right.

Because they’re usually faster than searching.

What it all Means

The joke about engineers forgetting how to code misses the point.

Our solution to being unable to use AI was to switch to another AI.

Which is no solution at all, but we actually need to look where we are before we all go shooting our mouths off at the situation.

The truth is not that we have forgotten how to code. That’s not even on the horizon.

The issue is that we’ve outsourced deep thought. I thought that AI would free us up from Kahnemann’s rather immediate “system 1” thinking and allow us all to think at a deeper level. I don’t see that as something that has happened, in fact quite the contrary.

When decisions are made, they are made in a rush. We decide to do things without considering the consequences, without really thinking about the follow-through. We “just” get things working, we think about the best way to push out features without thinking about the future maintenance of what we do. Which is what we always did, so I think there is no problem with our collective memory.

Yet we have doubled down on outsourcing the easy stuff. The for loops, the SQL functions and the underlying knowledge are all gone. We use tools for the basics and have a gap where the understanding never was, and seems never will be.

The moral of the story is that we haven’t forgotten arithmetic because we use calculators. It’s that most people didn’t learn arithmetic, and they continue to fail to do so.

When everyone has forgotten how to think nobody puts value in doing so, and that is dangerous in itself. If every design decision, architecture discussion and debugging strategy starts with “Ask Claude,” then eventually we’ll stop asking ourselves any questions at all.

It’s not a skill we will be able to build in the middle of an outage.

About The Author

Professional Software Developer “The Secret Developer” can be found on Twitter @TheSDeveloper and regularly publishes articles through Medium.com.

The Secret Developer believes AI is a fantastic assistant, but begins to see that they’re more expensive than a competent developer.