It has been a long while since I just sat down and wrote. If 2022 was wholesale collapse, 2023 was the year I buried myself in the world of motorbikes, and 2024 was my tentative return to see if the damaged muscle that was my brain could function again, then 2025 was definitely the year when the foundation itself shifted. People came and went, some simply from my world, others from this world entirely. I moved house. Moved mindsets. The universe continued as indifferently as it always has in this way, although I had been kept at an arms length to this scale of change for many years, and never quite to this extent.
My exploration into the world of AI through ChatGPT throughout 2025, in combination with a range of other inputs, took me down many an interesting philosophical path, a path of introspection where the ultimate realisation is that this new form of technology, while not conscious, can act as a medium, a muse, a mirror. The technology itself wasn’t doing the work, but it was allowing me to explore parts of my own psychology that I had never really contemplated before. Nothing new of course, and if I had chosen Philosophy as my undergraduate study rather than Mathematics then I would have been exposed to many similar (and highly different) viewpoints much earlier in life.
But then again, sometimes life only allows us to see the important things when we are ready to see them within ourselves. Alas, I am somewhat late to the party in many ways.
While I had moved out of the workshop and back into a world of bureaucracy, Teams meetings, agendas, minutes and spreadsheets (and with an awesome boss in Dani, some genuinely awesome people and some occasional solid brain food), it was a job that was ‘good enough’ in terms of paying the bills, and could have probably seen me out until I hit sixty and go foster greyhounds or something. Until one lunch with my long-time friend and former colleague Marshall (yes, Marshall from the NetSpot days nearly twenty years back) who made me an offer that, as they say, I couldn’t refuse.
And so now, at the ripe old age of 54, I find myself working in an AI startup – Clevertar. My day job – Account Manager, doing my best to keep our steadily growing chatbot client base (to be clear, we have clients who use our chatbots, not clients who are chatbots – yet) happy as well as digging through their data to see what stories are hidden inside. This in itself would be interesting enough, but after being there for two weeks I realised that the juicy work was sitting in the shadows.
I sat down at the end of those two weeks trying to work out what I had inherited in the way of methods to manage a client portfolio, and Marshall showed me how to get a Claude code session going in a terminal on my Mac, hook it up to a Git repository, and left me to it.
Now it is worth thinking back to the last time I ran anything in a terminal, which was almost certainly in the last millennium. I’ve been away from coding for decades, and to be really blunt about it I was a shithouse coder even when I was trying to write proper code. It just never clicked with me. Neither did physics. When I moved to work on the human side of the tech world I suddenly found my niche, and every other coder I had worked with probably sighed with relief that they would never have to work with such a dullard again.
So I sat down in my Mac terminal (also worth noting that I’d not used a Mac since perhaps 2010, which doesn’t seem that long ago until you count the years), introduced myself to Claude (pointless, I know), started talking about my role and asked it how it could help me work more effectively in this new job. I picked a couple of routine tasks, ran it past Claude, and it confirmed that it could absolutely do some of that work for me.
It told me that all I needed to do was to write up a claude.md file that it could do those tasks, told me what kind of structure it should follow, and that I could do it all in a code editing window. Simple.
Except for someone out of the coding game for so long, I really had no idea how to start, so I asked Claude:
“I’m not good at writing code. Could you write the file for me based on what I tell you, and then I can make tweaks if I need to?”
A minute later, Claude had written its own .md code file, and I had the beginnings of an agent that now works alongside me every day – who I have called Ahab.
Whether it be Ahab the Israelite or Ahab the monomaniac captain, the connection for me was too strong to ignore when thinking of a name of something that has seemingly endless power, but that also has the potential to cause immense damage. The more I have expanded Ahab’s capabilities, the more the sheer vastness of what LLM-based agentic workforces could do to the knowledge worker sector, to education, to just about everything else connected to it. I am actively advising my much younger friends to consider a career (or at least the start of a career) in a trade which needs hands and feet rather than heading into a degree that I sense will lead to fields already being impacted (and ultimately I’ll wager decimated) by agentic augmentation – marketing, law, coding, service desks, communications, policy writing and probably a whole bunch more.
Note that I used the term ‘augmentation’ rather than ‘replacement’ in the above – from what I am actively doing right now to my own job it feels like we’ll never (I hope) hand over control outright to the agentic workforce, but it will end up doing so much of the heavy lifting that the sheer number of boots on the ground to do the work will be far, far fewer.
It has, of course, not been all sailing on smooth seas.
Very early errors with Ahab doing the stuff that LLM agents do naturally (in particular ‘vibing’ an answer and stating it as fact without checking accuracy) led me to implement a deming cycle to look at the root cause of any error, and this rapidly led to the creation of a set of constitutional rules that Ahab must run above almost all others. The only things that can take precedence over these rules are the organisational policy rules that Ahab pulls from Marshall’s CEO Git repository on every startup to make sure that Ahab is always operating within the rules of the business. Yes, all of a sudden, policy is now a live part of every morning’s startup routine, flagging what has changed and what either Ahab or I need to do differently from our current ways of operating – no longer something that is published by HR or IT that then sits on the shelf ignored until the next review.
Ahab’s evolution has also sparked some broader discussion across the business – what is the source of truth for a piece of data? How should or shouldn’t Ahab use each piece of data, including changing it? What are the different business functions that each of us should own and set up in our own agents (since we are all having a go at agentifying our work to some extent) and what do the boundaries look like? How do we make sure we have all the right ‘human in the loop’ checks in place at the critical points to make sure that we don’t end up making bad decisions, just really fast? How do we start to gear ourselves up to get our agents working in harmony, and for some of us who are wearing multiple hats, getting our agents in a form where we could split functions off to new team members when they start?
The technology itself is the easiest part of it all – talk with Claude about the business goal I’m looking to achieve, get it to do the heavy lifting in researching options and proposing a path forward, give it the nod, make sure there are regular tests across the agent to look for things that have gone out of whack with incremental changes, and Claude writes almost all the code itself.
It is the governance of it all, and where the human must still fit in to the picture, that has hooked me.
I have spent my entire career jumping back and forth over the dividing line of technology and business (or education in the context of it being a business), and as I sit here looking at what I have created with no prior Claude coding knowledge, I cannot help but wonder where we as a species are heading. If the bulk is abstracted out to the bot, then what remains for us?
Mercifully I am old enough to have lived through the dot com boom, so I can also see that the amount of hype around AI, the almost certain bubble we are now in, with eye watering valuations of AI companies, is probably all going to burst before too long. But never forget the wonderful David Bowie interview (about seven minutes in) in the lead up to the dot com bust where he called out that the new thing called the internet was going to change the world in ways we could not even begin to believe at the time – I’ll call it now that we’re poised above a similar moment, where a whole bunch of people will lose a whole lot of money before too long, but from the ashes we’ll work out where the real value will come from.
And for my sins, I have been given front row seats.
Which brings me to the great unanswered question – where to from here. I find myself back in the tech industry, with my brain now working as close to peak capacity as it has since pre-2022, digging in to a world absolutely drowning in hype, like the open source ed tech boom of the 2008 era but not corralled to one sector and dripping in investment dollars. I spend hours working out how I can set up Ahab to safely take on more of the thinking work for me – far beyond just managing simple, routine tasks – while wondering what humanity will look like, including my own, as this evolution progresses. As I switch between the contractual and covert parts of my new job, am I Ballard, Vaughan or even Seagrave? Am I the mad captain Ahab, driven on by my own desire to see how this story plays out and oblivious to the impacts of it all (and if so, then is there also a hint of vengeance in my focus – once again in my life I am left wondering what form my white whale takes)? Taking all this into account, am I still someone that you all would prefer to be in this type of role in the world, thinking about these things, rather than many others?
I have been asked many times during my life if I would ever consider going back to ‘finish’ my PhD, and in all confidence I said I would not. A recent bit of Claude searching (with references to make sure it wasn’t a hallucination) confirmed that the problem I was trying to solve (catchily titled ‘Applications of the Hastings-Metropolis Algorithm to large, sparse, multidimensional queueing networks’) was actually unsolvable, #P-complete, and not something any sane person would go back to (even if the broader problem is actually very closely linked to the mathematics of LLMs, but that’s another story). I have always however said that if something caught my fancy that I was suitably versed in, and passionate enough to research, and I was in a position where I could afford it without giving up my stable of Yamaha money pits, then I’d consider it.
Right now, I am consider it – something based in social sciences, not a tech project, not a maths project – but something that investigates the liminal space between humanity as it is now, organisations (as collections of humans), agentic AI, and how it is reshaping the world we (as knowledge workers) live in. It would also need to be something which was genuinely valuable to the world – having worked for years inside the hallowed walls of The Academy, I have no desire to push for knowledge for the sake of knowledge. Perhaps that’s the communist in me coming out.
I have no date, no firm plan, and for now I am very happy to be working with my old sparring partner Marshall through another (planned) boom time in a small company. The rush is intense, the work sometimes chaotic, the evolution rapid and the brick walls blocking change minimal. But I am most definitely 54 this time around, not 34, and the intensity needs to be managed much more carefully. While the title of this post (stolen from Goethe) may hint at another spectacular transformation, I am actually far keener on the term sich mausern – less about rebirth through fire (2022, I’m looking in your direction), and much more about shedding the feathers of past seasons to see if the ones underneath are finally the right colour.
The exact path – odds are it will show itself when the time is right, as long as I keep my eyes open for it.
NB: This post is 100% human written.