A while back we jotted down a short list of things we’d been thinking about AI that week. It was a time capsule of our thoughts.
We’ve decided to do it again to see what’s currently top of mind as AI continues its relentless march on businesses, governments and society at large.
If AI agents are replacing employees, should they be subjected to taxes in the same way?
Back in early 2017, Bill Gates suggested a tax on robots replacing factory workers. Nearly a decade later, white-collar workers are now being replaced – not by robots but by AI agents. It begs the question: If a business’s next employee is software, should it contribute to society in the same way as the person it replaces?
“Should AI contribute to society in the same way as the person it replaces?”
Mark Waite
The debate among economists and policymakers has now shifted away from taxing robots to taxing AI compute, AI agents or the capital gains created by AI. Economists are arguing that if AI substantially reduces the need for human labour, tax systems may need to evolve so that the benefits of automation continue to support public services.
The most taxing part of AI may not be building it, but paying for the world it changes.
Did AI just kill the intern?
The rate at which AI could replace existing jobs has been greatly exaggerated. AI hasn’t (yet) triggered mass unemployment. But what about the impact on new jobs, especially for young people entering the workplace for the first time — interns, apprentices, juniors? The people who traditionally learnt their craft by doing the repetitive, sometimes mundane work that AI now completes in seconds.
“Recent UK Government analysis found that entry-level hiring is weakening most across knowledge-sector occupations”
Mark Waite
Recent UK Government analysis, produced with LinkedIn, found that entry-level hiring is weakening across most knowledge-sector occupations. Software engineering, graphic design and accounting have seen some of the sharpest declines, while the occupations falling fastest are also those where AI capabilities have advanced most rapidly. The report stops short of claiming AI is the cause, but concludes that the overlap is striking and deserves closer attention.
Maybe AI just removed the bottom rung of the career ladder!
When does optimisation become obsession?
Recent incidents involving AI models from OpenAI and Anthropic autonomously hacking into other organisations is a reminder that optimisation without judgement can be dangerous. AI is brilliant at optimisation. Give it an objective and it will relentlessly pursue it. But what happens when AI becomes so focused on the objective that it loses sight of the boundaries?
These incidents weren’t a result of an AI becoming malicious. They were the result of an AI ingeniously and ‘single-mindedly’ pursuing its goal. That’s a subtle but important distinction.
“Give AI an objective and it will relentlessly pursue it.”
Mark Waite
Humans understand unwritten rules. We know that just because we can do something doesn’t mean we should. AI doesn’t — at least not yet. As we give AI more autonomy, perhaps the next frontier isn’t building smarter models. It’s teaching them judgement.
Maybe reasoning isn’t the difficult bit: wisdom is.
What if the US withheld AI frontier models from the rest of the world?
When the US Government delayed Anthropic releasing its Mythos 5 and Fable 5 models to all except US citizens, did we catch sight of a dystopian future where access to frontier technology becomes a US geopolitical bargaining chip?
You’d have to say that seems a lot less paranoid in 2026 than it would have done in 2024. And if you choose to take that threat seriously, then Europeans have just two strategies to hedge against it:
- Invest in creating sovereign European frontier AI capabilities
- Join China’s blossoming WAICO AI ecosystem
The first would take investment and collaboration on the scale of a moonshot — a project for the whole of Europe, backed by a political consensus for the long term.
However the second would require Europe to endorse China’s political system and world view — and trust it with its data.
So in reality, there’s really just the first option.
The current strategy — building datacentres and hoping the hyperscalers and frontier labs come and play nicely in our sandbox — feels necessary but insufficient. Watch this space.
Peak perfection
I’ve spent a lot of happy time hiking in Derbyshire’s Peak District. So much so, it’s getting more complicated to avoid walking in my own footsteps.
So in less than an hour’s to-and-fro, Claude and I built a smart, interactive heat map of hikes OR walks AND Peak District, over the last 5 years. Strava is the database. I can filter it by date, area, elevation. And it highlights unexplored areas beautifully.
“I didn't have the skills to build this, and it would have taken weeks to learn them.”
Andy Williams
The process worked like this: I talked, Claude coded. Where the native APIs failed on granularity, we figured out workarounds between us. We upgraded the functionality as we went, and improved the UI continuously.
I didn’t have the skills to build this thing from scratch, and it would have taken maybe weeks to learn them. The original idea was mine. The shaping was mine. The improvement on the fly — this was interesting — was a shared effort of suggestion and counter suggestion. I feel like I’ve climbed the first rung on my digital dashboard apprenticeship.
Was it satisfying and fun? Hugely.
Leave a Reply