The following piece was submitted to The Spectator, January 2026
The feeling of renewal that comes with New Year often dissipates as the return to work hoves into view. January blues set in. This time around, though, I’ve been unable to shake an ennui not brought on by middle age, my daughter heading back to university, or even signing into Teams.
No, this malaise was brought on by using a specialist tool for programming provided by Anthropic (they of the ‘Claude’ series of large language models and an OpenAI competitor). Their ‘Claude Code’ software is like having an expert sat at your desk, hands on your keyboard. It can see your files, read your error messages, make changes, run them, see what breaks, fix it, and iterate - all without you playing messenger. It’s a key tool in what’s been dubbed ‘vibe coding’. You describe in English what you want to happen; it does it, in your actual environment, handling the how and showing you the result. Run several at once and you have a team of ‘agents’ - the term of art - each tackling a different problem.
With older versions, this tool was great fun to use but would often produce unintentionally hilarious (to programmers, at least) results or get itself stuck in logical cul-de-sacs, unable to reverse course without wrecking virtual flowerbeds. Since the launch of “Opus 4.5”, Anthropic’s latest and greatest model, something feels different.
It has gotten way, way better. I’ve had a list of projects in my head (for several years in some cases), personal little tools or tchotchke-like creations I could never make any progress on. The damn thing has ploughed through them all, causing me to stare at my screen like a Victorian at a séance who’d finally got a genuine reply.
Others have begun coordinating teams of agents to tackle larger problems and managing them like virtual employees. It’s like playing at being Mark Zuckerberg where the agents are your young frat bro programmers, only without the need for free office food, beer pong tables or actual sleep. Businesses are being launched. Talk is rife about the first solo founder billion-dollar company where the staff are all agents.
FOMO had me forking out for higher subscription levels. At some point my wife is going to notice. I’m not sure what spending threshold will see us heading to marriage counselling. Perhaps the agents could figure that out. To win her over I plan to do what others have done, introducing the agents into the family WhatsApp group, slowly making them indispensable members of the household, domestic staff checking our calendars and remembering our shopping lists.
But back to the malaise. This was no ordinary January slump but rather a realisation that I’d glimpsed what work could be like with friction removed and tedious bits automated. I imagine this is what a rural visitor to the Great Exhibition experienced, going home to his hand tools after seeing a steam engine do in seconds what usually took him hours.
Back at work, the team I’m assigned to had its usual morning meeting, picking up the flotsam and jetsam of the post-festive work queue. Tasks were allotted conservative timescales and I sat there knowing we could have them all done by lunch if the new tools were available to us. I imagined just throwing these at the agent horde, ever hungry for tasks. It’s not that AI is wholly banned at this particular job either; it’s just that the cutting edge takes a while to arrive in corporate settings. I’ve watched this happen before.
Of course, my silly projects are not huge governmental systems nor infrastructure relied upon by financial behemoths. There is a world of difference, but the principles are the same. There is work to be done and code to be written, and the agents heading our way will do it all. My LinkedIn feed is awash with reassuring posts about how AI will never replace human creativity, human judgment, human something. I used to find them comforting. But for now, I pick my hand tools up and whittle away.