A lot of AI news arrives each week sounding bigger than it is. I wanted to start writing down the few stories that actually change how we should think about skills, adoption and growth in the UK, and what they mean in practice, not on a keynote slide.
The entry-level jobs warning keeps coming back. The work is what we do with it
Anthropic's Dario Amodei was in the news again this week for a warning he first made last year: that AI could wipe out half of all entry-level white-collar jobs, in fields such as law, consulting and finance, within one to five years. (Axios)
He is describing a transition, not a verdict. Every time technology moves on, some old entry points close and new ones open that nobody has named yet. The real question is whether the people coming into the job market have the skills to step into those roles. That is the work we do at AI Lab. We've now taken nearly 1,000 students through elements of the Zygens framework. They work on real problems, build their first agents, and leave with skills most of the people already in those businesses do not have yet. Those are the conversations we should be having: not only what disappears, but what gets built on the other side. I wrote more about this in “The entry-level jobs aren't disappearing. They're moving.”
Yorkshire just became part of the infrastructure argument
On 3 November, Drax, Brookfield, Radiant and Qai announced plans for a phased AI data centre at Drax Power Station in North Yorkshire. The first phase would be up to 100MW, aimed at operating from the first quarter of 2028, with initial backing of £2.5 billion. The longer ambition is 500MW between 2028 and 2031, and potentially more than 1GW in later phases, with total investment that Brookfield estimates could exceed £20 billion. It is still a plan. The first phase depends on planning permission, regulatory and network approvals, and finding a tenant. (Drax announcement)
This is a Yorkshire story, not just an energy one. If it happens, it is about whether the North hosts some of the infrastructure the UK says it needs for sovereign AI, and whether local people and businesses actually benefit. Compute on our doorstep does not automatically mean capability in our firms and colleges. That part still has to be built: the skills, the use cases, and a say in what the investment is for. A public consultation has since opened, running until 3 December 2026.
“Let everyone experiment” is not a governance model
Bloomberg reported that Goldman Sachs has restricted Anthropic's Claude Code and Microsoft's GitHub coding tools to its engineers, after months in which bankers and traders were also allowed to use them. The bank pointed to a review of productivity and risk. Specialist “strats” are exempt. A spokesperson said that after a period of experimentation, the firm puts tools “in the hands of those who are best suited to leverage them for the long term.” (Bloomberg)
This is a pattern we see in organisations of every size, not just banks. You open the tools up, people get excited, and then someone asks who is allowed to use what, on which data, and who owns the output. That question arriving late is what turns a useful experiment into a mess. The answer is not to ban everything. It is to decide, early, which roles get which tools, and who checks the work. If a global bank is tightening access after the experiment, smaller UK firms should not skip the experiment, but they should not skip the rules either.
The best agents still fail most of the real work
On 2 November, SemiAnalysis released GDPx, a benchmark of 206 realistic professional tasks across 11 domains, with agents working for up to eight hours in proper software rather than answering a prompt. The leading model, Claude Opus 5.5, scored 12%. (GDPx)
This is the most useful kind of disappointing result. It does not mean agents are useless. It means they are not ready to be left alone on high-stakes work. In our labs we keep coming back to the same point: stay in the driving seat, check the important bits, and wrap the model in a process a person can stand behind. The scarce skill is not prompting. It is knowing where a person still has to own the outcome.
If you want to take this somewhere practical
If you want a plain read on where your own organisation sits, take the free AI Lab Benchmark: 15 questions, about five minutes, and a practical next step. And if you would rather practise this in a room than on a screen, upcoming sessions are on our events page.
The next Leeds date is Tuesday 1 December at Platform, Bruntwood SciTech: AI Clinics (9–11am), AI for Business: Lunch & Learn (12–1pm) and Find Your First AI Agent (2–4pm). These sessions are run for Bruntwood SciTech customers and ecosystem.
About the author
Charlie Bartle
Co-founder, AI Lab UK · CIO, Zygens
Charlie co-founded AI Lab UK and runs the lab programme on the ground. She works with SMEs, councils and universities to turn AI curiosity into working tools people actually use.
