June: Who doesn’t love breakfast?

We’ve had the full spectrum of weather recently here in London, and we’re basically British now so we’re prepared to talk ad nauseum about it. Only joking, what we’re really talking about is our first ever Friday Breakfast Club.

Just 10 or so like minded folk who are excited by data and want to hear one of the OGs, Charlie Green, founder of RFI Global, talk about how to build a very successful business, and what he'd do differently in hindsight. If you’re in London on the 26th June and want to come for a very good coffee, email us, because spots are strictly limited.

We've also launched our very first Instagram account. It's very much a work in progress, but if you want behind the scenes activity, reposted reels and an alarming number of dog photos, that's the place to be.

And now, some news that’s not about us. This month, Europe is worrying about America's digital 'kill switch' and Amazon has discovered that measuring AI usage is a terrible KPI.

Dependency Day

A recent FT Big Read uses the case of ICC judge Nicolas Guillou, sanctioned by the US in August 2025 over arrest warrants for Netanyahu and Gallant, to illustrate how exposed Europe is to American digital infrastructure. Within days of the sanctions, Guillou lost access to credit cards, Booking.com, Expedia, UPS deliveries, even Paris's Vélib' bike scheme, because compliance departments across these services refused to take any risk on US enforcement.

European officials now describe this as the 'kill switch' problem. That is Washington's ability to disrupt everyday life far beyond its borders because of Europe's reliance on US technology and payment systems. What once sounded far-fetched has become a live policy concern.

Europe is responding with plans for alternative payment infrastructure, investment in cloud and AI, and renewed discussion about digital sovereignty. The challenge is that demand for US technology remains incredibly high, and AI may deepen rather than reduce that dependence. The uncomfortable reality is that most European businesses already depend on US infrastructure so replacing it is easier to discuss than deliver.

Lobster Management

The New York Times profiles small-business owners using software called OpenClaw to create armies of AI agents, affectionately known as 'lobsters'. These agents answer customer enquiries, manage bookings, analyse documents, run advertising campaigns and even spend money on behalf of their owners.

The results are impressive. One entrepreneur reduced 20 hours of monthly admin work to around an hour. But there are downsides. Agents sometimes delete emails, leak information, hallucinate instructions and generally behave like the worst employee you've ever hired.

What stood out to us is that the businesses succeeding with AI agents all ended up building governance around them. Turns out even virtual employees need supervision.

Cheezus, that's a saving

A Mark Cuban-backed vegan cheese company found one of the least glamorous uses of AI imaginable: checking cardboard boxes. Rebel Cheese built an agent to compare shipping invoices against contracts and spot tiny packaging issues that triggered carrier surcharges. The result? Around $400,000 in savings.

We love this because it's really a story about unexpected data value. Rebel Cheese wasn't sitting on a breakthrough algorithm. It was sitting on years of invoices, contracts and operational data that nobody had connected properly. Sometimes the biggest opportunity isn't generating something new. It's finally making use of information you already have.

Running on empty

Amazon has shut down an internal AI leaderboard after employees started gaming the system. The tool ranked developers based on their use of Amazon's AI coding platform, but instead of encouraging productivity it encouraged "tokenmaxxing", assigning AI agents pointless tasks simply to drive up scores. The result was higher computing costs, lots of AI activity, and not much evidence of useful work. Amazon's message to staff was direct: stop using AI for the sake of using AI and focus on building better products.

The lesson here is that the second you turn AI adoption into a KPI, people optimise for that. It's the same reason sales teams chase vanity metrics and executives obsess over slide counts. So if you're going to set AI KPIs, maybe make sure you're measuring the right thing.

The Report Card

This New York Times piece tells the story of how Khan Academy and OpenAI built Khanmigo, one of the first large-scale AI tutors. Sal Khan initially rejected OpenAI's advances, but then he saw GPT-4 explain a concept, generate a question, and personalise feedback in real time, and changed his mind fairly quickly.

What follows is a fascinating account of the messy reality behind successful AI products. GPT-4 hallucinated, struggled with maths, displayed biases and frequently forgot its role as a tutor. Much of the work involved prompt engineering, testing and repeatedly refining behaviour rather than writing code. The teams had to decide how an AI tutor should respond to sensitive topics, cheating, mental health concerns and historical subjects.

The takeaway is anticlimactic. Khanmigo is a useful product that took years to build, student engagement was mixed, teachers often found more value than students, and the technology still required extensive human oversight. Which is probably what we should expect. We'd be worried if the story ended with everyone declaring AI had solved education.

The consultants' dilemma

The Financial Times explores whether AI could do to consulting what the internet did to newspapers: remove the advantages that protected incumbents for decades. New AI-native firms are emerging across the UK, often founded by former Big Four partners who believe AI agents allow small teams to compete with organisations that previously relied on thousands of junior consultants.

The article identifies three pillars of the traditional consulting model that AI threatens. First, clients increasingly use AI themselves for initial research, reducing demand for generalist consultants. Second, AI undermines the billable-hour model because tasks that once took days can now take minutes. Third, it weakens the pyramid staffing structure where large numbers of junior staff generate profits for a smaller group of partners.

The interesting nuance is that the FT isn't predicting the collapse of McKinsey, Deloitte or Accenture. Instead, it suggests the market is fragmenting. Boutique firms become more competitive, specialists become more valuable and pricing shifts towards outcomes rather than hours worked. For anyone running a consultancy, the article feels less like a prediction and more like an early warning. Bit of a whoopsie from us then..

The Maccas Millionaires

While everyone is talking about AI replacing white-collar work, The Economist makes an unexpected case for burger restaurants. Franchises now account for roughly one in eight American businesses with employees and have quietly created enormous amounts of wealth.

The article argues that AI may actually make franchise ownership more attractive, not less. If software can increasingly perform knowledge work, then businesses rooted in physical locations, local relationships and thousands of small operational decisions start looking surprisingly valuable.

It's a slightly uncomfortable reminder for anyone obsessed with technology. The future may belong to AI, but plenty of fortunes will still be made selling burgers, changing tyres and cleaning swimming pools.

More code, less value

The debate around AI productivity is maturing. Nobody seriously argues that AI can't increase output anymore. The question is whether that output translates into value. A new study found that developers using AI edited almost 300% more files, but the increase shrank dramatically as work moved through reviews, approvals and release processes. By the time software actually reached customers, the gain was closer to 30%.

That's still meaningful. But it helps explain why so many organisations feel simultaneously impressed and disappointed by AI. More work is happening, but the bottlenecks simply moved elsewhere. The comparison to electricity is a good one. Productivity didn't explode when factories installed electric motors. It came when they redesigned the factory around them.

Watercooler Chat

A section of the things we like that keep us sane while running a small business…

• Palace of Parliament Romania: The world's heaviest and second-largest building, covering 365,000 m² and with 2.5 million tonnes of marble.

• Salus: The lifting gym under our office, Emma and Lauren are both obsessed, and have actual muscles now.

• Vesper: Jackson Boxer’s new restaurant in Exmouth Market is fantastic. Order a sgroppino and the salt and vinegar crisps (trust us).

• Slopaganda: This crazy FT article about Donald Trump’s meme machine. It's both awe inspiring and appalling.