On LLMs, Push Joints, and embracing the change

2025-05-29

I tried to do some plumbing once. It’s a cautionary tale about capability blind spots and the all-permeating power of water. Eventually, the repairs to the walls and floor far exceeded the money saved by doing it myself.

I remember talking to one of the plumbers who came. He didn’t seem too perturbed about my lack of skills, I had the impression he had seen it before. Almost as an aside, he told me briefly about his dad, also a plumber, and push joints.

When push joints came in, so he said, many plumbers thought that was the end. Who’d need welding or pipe bending, which were after all the hardest parts of the job? The plumber’s dad stared gloomily at his own, inevitable obsolescence.

Or so it seemed. Fast forward several decades, and it appears we still have plumbers. Not just to repair the damage caused by gung-ho pseudotradies who think B&Q holds all the answers.

You see where I’m going with this. We are currently, absolutely, and without a doubt, in the middle of a major change in how technology is built, delivered and used.

Large language models are not even done. Models are improving, memory capacities are growing, capabilities are extending. And beyond that, all that was machine learning has had a major shot in the arm.

Technology doesn’t come more disruptive than this. And yes, there is every indication that AI is going to take your job. But, we’ve been here before, several times. I know I have.

First, the bad news. The jobs that appeared stable now appear anything but. Nobody is immune, even senior roles who were remunerated on the basis of their inherent knowledge and experience.

We’ve been here before. Back in the Nineties, I watched as people whose careers were built around maintaining proprietary software systems, saw them replaced by commercial, then open-source stacks.

Later, we can all blame Tim Berners-Lee for instigating what became known as disintermediation. The web was astonishingly disruptive to supply chains, taking away livelihoods even as it created new opportunities.

In neither case, nor in legions of others, were people happy. We can focus on that, or on the fact that employment is still a thing, despite Seventies predictions suggesting we’d all be living lives of leisure by now.

We still need people to do things, just as we always did. Will they be the same things? No; but neither is being a farrier a good career move.

As my lovely colleague Darrel Kent once put it, technology either automates, or it augments. If your work is automatable, it will be automated—this was always true, it was just a question of timing.

However, where technology exists to augment… that’s where the good news kicks in. Some tasks, not least being an analyst, benefit from the immense distillation power of machine learning models.

Equally there are certain places AI can’t go, and maybe never will. My money’s on hairdressing and baristas. It’s also on the outer edges of complexity, which is fractal. Every time we achieve more, we create more problems to solve.

See also: cybersecurity, which funnily enough also follows laws of permeability. Hackers find the easiest path to the money, they don’t care which route or which vehicle. Plus, they will always be one step ahead.

See also: areas of risk which require discernment. Whilst computers may currently be doing a great job at serving up answers, we often need a human in the loop to determine if the answers are correct and appropriate. That was true when “computers” were people.

As for innovation… I’ve likened this to mice dancing in a moving train. However good the underlying platforms may be, however fast moving they are, and however small we are, we can still dance on top. We can always do more.

Which is the very nature of business. Fundamentally, businesses have to do more, or different, than the competition to survive. They have to differentiate, to have additional capabilities, to have their own spin.

In every future scenario, this will remain true. Businesses stand between supply and demand, and take their cut. The clue’s in the word “enterprise”, whether at an individual or corporate level.

Will there be new intermediaries? Yes. Will there be charlatans and sharks? Yes. Will, again, there be jobs lost? Yes, but there will also be jobs we’ve never even heard of, alongside augmented versions of old jobs, alongside ones we still want people to do.

As a final point, the one thing AI still lacks is a meta-cortex, that is, the ability to review a bunch of disparate factors and discern an appropriate path. Just as complexity is fractal downwards, context is fractal upwards: it grows exponentially.

Perhaps “models of models” will exist in the future, but we are a long way off. See also, Artificial General Intelligence (AGI). Whatever this is, it is not that. As my other lovely colleague Whit Walters has said, LLMs are currently parlour games, not AGI.

As are AI’s use in creativity. I fundamentally believe art relates to the communication between two souls. Whilst AI music, or an AI picture, might appear similar to something a human can do, it lacks this fundamental characteristic.

AI doesn’t understand history or the ramifications of its actions: it can’t walk in the shoes of a humans, nor would it understand why without a prompt. It can’t make leaps of insight - it won’t sit, troubled, pondering an issue until it arrives at a sudden epiphany.

Perhaps one day it will be able to do this, and we will have our lives of leisure… I have an inkling that neither the best nor the worst intentions of our peers will allow for this scenario.

For now, it’s about embracing the change, potentially passing through the fires of a career move (sorry), but in recognition that, indeed, we are mice and we can dance. If push joints weren’t going to destroy us, we’ll be OK with AI.