Posted on
by

I have previously written about how AI is going to eat your lunch, and while I think that a lot of the hype around AI is unfounded, it is definitely not nothing. The problem that few seem to get is that we are at the bottom of an exponential curve in terms of capability, and exponentials are something the human brain really isn’t wired to understand.

We are at the start of this process, but it already has got a lot of people worried. We have been through large rounds of disruptive automation and outsourcing before, but unlike those former rounds which fell on the working classes (which nobody ever seems to care about when it happens), this time the axe is going to fall on middle class knowledge workers who had previously considered themselves immune. The first wave of disruption may arrive much sooner than people expect, perhaps within a few years rather than decades. Certainly, if I was an accountant, contract lawyer, or doing diagnostic medicine, I would be … very worried.

In my line of work, I have already seen significant disruption in my day to day and I have previously written about how AI has effectively become another team member to which I delegate work to, for better or worse.

AI is not going to go away, and indeed it is only going to get more sophisticated, but if you think about it we have seen this before when it was humans disrupting things through outsourcing, and to survive you need to adapt rather than fight (or cope that “they couldn’t do my job”).

As far as I can see, you’ve got two option.

Don’t be a widget maker

First off is to accept that if you’ve previously been someone who crafts widgets (be that writing code, or crafting contracts, or even designing buildings or producing artwork) you’re going to have to shift your role. What I mean is that in the (near) future it’s the AI that is going to be crafting functions, writing clauses or holding the virtual pencil.

AI is going to handle the practical process. The How.

However, what AI can’t do, and likely never will in a way that’s useful to humans, is decide what to build and to verify that what has been built is what was wanted. The what and the why must always be driven by the human, so long as humans live, since it’s human wishes that we are seeking to make manifest.

So as a craftsman, artist or engineer, your role shifts from tradesman to CEO. You write the specification, clearly (which is an art in its own right), and then use your expertise to validate the results and offer changes. This also involves gathering those requirements from other squishy humans.

Of course it’s the last step that’s the thorny issue as AI progresses, since it’s only as junior engineers and journeymen that we get to understand what “correct” even looks like. AI makes mistakes, whether by hallucination or by misunderstanding of the spec (which happens with humans as well), and so you will need to check that what you get is what you want functionally, and in terms of the why.

What does this mean for you?

Understand that the value you’re going to provide is no longer going to be “slinging code” or similar. You’re not going to be paid to write code. Your client can do that themselves now. Your job is to understand the what and the why. To provide the strategy, and understand what needs to be done to get there. You’re not going to be paid to write functions and classes, in the same way nobody is paid to copy books by hand.

You need to get to the point where you’re thinking in terms of delegating work to others, and writing clear specifications you can hand over to a human rfn, and get comfortable doing code review. As a junior dev this might be hard, but get there somehow. If you’re not using the AI tools available for your industry, stop what you’re doing and do so now. Speaking directly to my fellow introverted software engineers, I’m afraid you’re going to have to work on those social skills as well.

The safest place to be in an AI economy is not producing the work, but defining and judging the work.

Trade on luxury

Harder for an engineer, but easier for artists I suppose, but I will point out that people still pay money for artisan furniture, overpriced sourdough and pictures by famous painters, even though there are cheap mass produced options available.

Hand built furniture or Ikea. Both markets have customers, but you have to pick one. If AI makes production cheap, the remaining value comes from trust, taste, reputation and experience.

Do you have enough cachet that people will seek you out rather than others, and pay a premium for it? Harder, but I’ve made a living out of that even in IT, since through my involvement in certain niche projects meant I was the go-to guy for implementations. People would pay a premium for my involvement, development, and to my first point, strategic advice.

You’re going to have to sell the whole experience, which means branding, sales and after care (again, squishy human stuff). Apple, famously sell the entire experience of buying their products – from store, to packaging, to technical integration. People pay a premium for that, even in a world where there are technically better computers and phones.

They trade on the status not the device. The iPhone is marketed as the high status phone, in the same way as both an Limo and a cheap Honda will both get you to your destination just as well, but you’re unlikely to see Taylor Swift turn up to the Grammys in the Honda.

Be high status.

Inevitability

Whether it’s overhyped or not, AI is coming, and it would be wise to plan for that.

This shift will break a lot of old assumptions about work, but it also creates opportunity. The tools we already have dramatically reduce the cost of experimentation, and in an AI world the winners won’t necessarily be the best craftsmen, they’ll be the fastest at trying ideas.

The people who adapt will do very well.

Everyone else will be competing with the machine.

So protect yourself. Start experimenting.

..and maybe learn to vibe code.

Because if you don’t, someone who did will replace you.

Posted on
by

I made a thingy.

Or rather, I vibe coded a thingy. Essentially, I got bored during a meeting and rather than chew on my pen, I spent half an hour putting together a very simple monitoring server for our estate in order to power our status page, as well as to hopefully syndicate to the wider EOSC Node infrastructure.

I was aware that tools like this exist, but those all seemed needlessly complex for what I was after, and so I bashed together a quick node server. Tools like ARGO are fine, and are very capable, but aren’t containerised currently and require heavy weight things like backend databases and stuff. Not what I was after.

Anyway, this server is fed a config.json that lists all the server groups in your estate, endpoints and poll periods, and then exposes their status via API. This is then picked up by a javascript front end running in an S3 bucket.

What’s notable is that I didn’t actually write a single line of code here, from scratch to deployment I described what I wanted to the AI as if I was talking to one of my junior developers, and verified its working as I would a series of pull requests.

I’ve got a blog post coming about how I feel about this, and what I think it means career wise, but for now, enjoy this new (hopefully useful) tool.

» Visit the project on Gitlab...

Posted on
by

“Was du ererbt von deinen Vätern hast, erwirb es, um es zu besitzen.“

I have to confess that these days I rarely write code by hand.

Whether in my consultancy business or at my day job as Head of IT, when I actually have to sit in front of Visual Studio at my keyboard, I am mostly tweaking and debugging someone else’s code, or describing what I want rather than crafting it line by line.

Once, this code used to be entirely the product of squishy humans, but these days… not so much.

I write this sitting in Heathrow T5, waiting to fly to Hamburg for an internal meeting on the future of structural biology and data management. AI will dominate the discussion, as it now dominates almost every serious technical conversation. AlphaFold revolutionised structural biology, and now the entire field seems to be in a race to collect as much data as possible in order to feed the ever hungry models.

My mind wanders to software engineering, the industry I spent most of my professional life, and to my life in general, where this latest technology has entered the world and is already having a massive impact. Summoned by the genius of Man, like a real world Mephistopheles, providing easy knowledge to all who but ask.

I think on how much my work is changing, from writing lines of code to orchestrating and debugging. Shifting from the how, to concentrate on the what.

In the past, I would have to hold the entire model of a software application in my head. Crafting its aspects line by line, in the same way as a craftsman might build a building. Every creation bespoke. These days, I mostly describe my vision to others, I concern myself with the what and the why, and leave the how to another. AI, much like a team of junior coders, is leverage, and allows me to generate plausible looking software very quickly (insanely quickly with AI), and so my role has shifted from craftsman to architect. Specifying intent clearly, detecting subtle errors and integrating outputs into real live complex systems.

The irony of course being that the less I type, the more responsibility I have.

Debugging the Alien Mind

The mind of an AI is not human, and it can be wrong in subtle ways that even an experienced engineer has a hard time spotting.

It cannot be trusted in the way a human author can, it hallucinates confidently and fails silently.

There are rewards for those of experience who can wield this new tool, and I personally have found it to be a major source of incredible leverage, making my role faster and more efficient, and leaving me to focus on the more impactful aspects of my role. But the necessary ability to spot the subtle errors and misunderstandings quietly punishes newcomers.

Do we comprehend the deal that we are making?

Pulling up the ladder

Senior engineers come from being junior engineers, and as AI eats the white collar world like industry previously came for blue collar, it is realistic to assume that in the near future there simply won’t be junior engineer roles anymore. Junior engineers used to learn by writing bad code. If AI removes that phase, we are not accelerating learning, we are removing it.

I am benefiting personally, but what about the wider world?

It’s a trap, of course. While I’m fully cognisant that I am training my replacement, and sharing data with fairly shady individuals, I have little choice. Individuals and business that don’t use AI, for whatever reason, will soon be out competed by those who do. Soon, much like how a smart phone went from expensive luxury to practical necessity, AI will no longer be optional… and I’m beginning to think that point is behind us.

Certainly, right now, if you work in any form of white collar or creative field I would make a priority of learning the capabilities of the tools available. You don’t have to be an expert, but these tools exist and AI is coming for you regardless, best be prepared.

Opportunities… but for Whom?

As I previously stated, AI is leverage.

Whatever your field, a one man band can now produce things that entire teams would be needed for even a short time ago. Some decry this as the dying of art, while others embrace the new tools and produce wonders (and having got deep in the world of AI generated music videos recently, I can tell you first hand how impressive these things are even in their infancy).

Solo developers now have to tools to produce working tools very quickly and with minimal cost. Debugging complex problems now becomes a rubber ducking therapy session with ChatGPT.

There is opportunity there, but I don’t think it will arrive evenly distributed. The winners will not be coders, it will be system thinkers. It will be the people who can conceive of what needs to exist that didn’t before, rather than those who can craft optimal C code.

Sadly, however… you need fewer of those people, and therein lies the rub. Once your local grocery store had rows of checkouts, now it’s mostly self service check outs and one bored guy there ready verify your age when you try and buy a beer. Humans are the most expensive line item of most businesses, and so there’s a strong incentive to have fewer of them, and what you do have being higher skilled.

Telling yourself that “AI couldn’t do my job” feels increasingly like denial. This is still the dial-up modem stage… and we know how that story ended.

Faust vs Doctor Faustus: Two Visions of the Bargain

In the story of Faust, Goethe offers us a tale of striving, doubt, and eventual redemption through effort and responsibility. While in Marlowe’s Doctor Faustus, we are offered a shortcut to power, followed by loss of agency and damnation.

In both tales, Mephistopheles is a shortcut to power and knowledge – astonishing capability with little upfront cost. But the question we are asked in both tales is what we give up? I rather think my industry, and the world in general, hasn’t made up its mind which version of this story we are living in.

On one hand we may face a world full of mass unemployment, deskilling of the population as we defer to AI for everything, concentration of power into a handful of individuals who control the models, and a world of fragile systems nobody truly understands.

On the other, we may head into a world where humans are freed to focus on what makes them uniquely them, with powerful tools that let them create uninhibited. As a kid, I often imagined what I could create if I, say, had access to a “do what I mean” AI and a Star Trek replicator. Could we be moving into that world?

Ultimately, we’re in the middle of the story, and anyone who claims to know what will happen is almost certainly wrong.

Choosing to Earn What We Inherit

Ultimately, AI is a tool. One that gives us access to incredible leverage, but without caution it could be very dangerous. A bargain with a demon that we don’t fully understand.

We can’t un-invent it, and given the growth in capability, we’re at the bottom of a hockey stick of exponential growth in power. The singularity? Probably not. But that doesn’t really matter at this point.

However, we can face this with wisdom. Judgement not blind automation. Something to work with not surrender to. Ultimately, will we earn this power and salvation, or will we be consumed by it?

I type less than I ever have, and yet I feel more responsible than at any point in my career. That, perhaps, is the real bargain we are making… not less work, but more consequence.