The industrial revolution of our time started in early 2025:

There’s a new kind of coding I call “vibe coding”, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.

— Andrej Karpathy

By the end of the year, the world was blown away by the trendy new thing that made software development suddenly accessible to everyone: a high-dimensional, parameterized, probabilistic function. A Large Language Model, LLM. Not a new technology per se, but this time, the idea was to give it great autonomy and let it rip with little or no supervision.

Good vibes!

It promised many things and the hope was real. People started using it, feeling empowered. Some with success. Others, not so much.

That’s because many people were sold a lie: LLM = AI.

An LLM looks bright, but it is not AI: Actually Intelligent. It isn’t capable of true reasoning, and it is not creative. It understands its current context. It has no idea of what its broader business implications are.

It takes all of its training data and uses advanced mathematics to find a mathematically plausible combination of tokens: the response. It works most of the time.

When an LLM makes a mistake, though, it makes it with assertive confidence. The type of mistake most difficult to recognise by an untrained eye. Over time, the problem grows at the speed of thousands of lines of code per hour.

Two wrongs don’t make it right, and thousands of wrongs make it a serious problem when it comes to software development.

If that’s your world right now, we encourage you to get in touch.

An LLM can be useful, but that’s not what happens automatically - not yet, anyway. An LLM needs precise guidance, guardrails and acceptance criteria.

We may be able to help you get out of the rabbit hole your LLM dug up for you.