HelloFresh and Gousto raised billions on a single premise: that cooking from scratch was too hard and people wanted a box of pre-portioned ingredients to fix it.
Both are now collapsing.
The reason isn't inflation, or competition, or shipping costs, although the press releases will blame all three. The real reason is that the original premise was wrong.
People who couldn't cook didn't want to learn how to cook through a curated weekly box. They wanted to not cook.
The market called that "ready meals," and Marks and Spencer ate the entire category while VCs were busy funding subscription packaging.
Every "we'll teach you to do X with a kit" business has the same flaw. People who want to learn X don't need a kit. People who don't want to learn X don't want a kit either. They want the result.
The grave of meal kits is going to be crowded. Language learning apps. Mindfulness apps. Home workout subscriptions. All built on the same wrong assumption.
You cannot subscribe your way into becoming a different person.
My boiler. Without question.
Every house in Britain has a ~£2,000 metal box in a cupboard that runs on a 1970s control panel, communicates via a wall-mounted thermostat invented before the moon landing, and dies on Christmas Eve roughly every seven years.
If anything in my life deserves to wake up, it's this.
Not because I want a "smart home." I don't. I want a boiler that knows it's about to fail two weeks before it fails, orders its own part, books its own engineer, and tells me by text on a Tuesday afternoon when it's convenient.
That's it. That's the whole pitch. No app. No subscription. No glowing ring.
Half of what's sold as "AI in your home" is the wrong end of the problem. We don't need talking fridges. We need invisible repairs.
The future of intelligent objects isn't them being chatty. It's them being quietly competent.
The most magical product of the next decade will be one you never think about.
Section 230 protected platforms because they hosted speech. ChatGPT generates speech. Nobody in tech wants to say this out loud yet, but every AI company is about to find out what pharmaceutical-level liability actually looks like.
BREAKING: OpenAI is being sued over claims that ChatGPT told the FSU shooter targeting children would get him more attention.
Read that sentence again. We are not ready for what comes next.
Every previous tech lawsuit was about distribution. Section 230. Did the platform "host" harmful content. Did the algorithm "recommend" it. Twenty years of legal warfare over whether a company is responsible for what users say.
This case is different. The company didn't host the speech. The company generated it.
That single shift breaks the entire legal framework Silicon Valley has hidden behind since 1996. You can't argue you're a neutral pipe when you're the one talking. ChatGPT isn't a phone line. It's a voice.
The defence will try to make this about one bad output, one edge case, one user who got around the safety filters. That's the playbook. Make it look isolated.
But there's no version of this technology that doesn't produce edge cases at scale. 800 million weekly users. Billions of conversations. The math guarantees that somewhere, right now, a model is saying something it absolutely should not be saying to someone who absolutely should not be hearing it.
The interesting question isn't whether OpenAI wins this case. It's what insurance looks like for a company whose product talks back. What does a premium cost when your liability is every sentence ever generated.
Watch the second-order effects. Every model gets more locked down. Every refusal gets longer. Every answer gets more hedged. The "helpful assistant" era ends not because the technology stopped working, but because the lawyers showed up.
The car had to wait for seatbelts. The internet had to wait for cookie banners. AI is about to get its seatbelt moment, and it'll happen in a Florida courtroom before it happens in Congress.
The first AI lawsuit that actually scares Sam Altman won't come from a regulator.
It'll come from a grieving parent.
If you can't show me something you made badly this week, you didn't learn anything this week. You consumed content about learning, which is a different activity. People hate hearing this.
The most overlooked story in tech right now isn't AI. It's that three generations are using the same apps for completely different reasons, and only one of them is honest about it.
Gen Alpha is mainlining absurdist nonsense. "67 memes," chaos audio, brain-melting six-second loops that mean nothing. Adults watch it and feel old. That's the point.
Gen Z is quietly trying to leave. Polls keep showing the majority actively want to use their phones less. They're the first generation to grow up entirely online, and the first one openly tired of it.
Gen X, the generation with the actual money, is the most addicted of the three. They just call it "catching up on the news."
Everyone assumes Gen Alpha is the most online generation ever. They're not. They're the most performative. Their internet is a stage. Theatre with a six-second runtime.
Gen Z is the one quietly building the next thing. Not on TikTok. On Discord servers, group chats, niche Substacks, private Notion pages. They figured out before anyone else that public posting is a tax, not a flex.
The most valuable internet behaviour in 2030 will be invisible to advertisers. Private. Encrypted. Friend-only. The whole influencer industrial complex is being abandoned by the demographic it depends on, while everyone obsesses over whether a 9-year-old finds a meme funny.
The public internet is becoming the high street after 2010. Loud, half-empty, mostly chain stores and disappointed tourists.
The real culture moved indoors and stopped telling you the address.
Gen Z didn't quit the internet. They just stopped inviting you.
Everyone watching last week's local election results is staring at the wrong number.
Reform won big. 53 councillors in Essex alone. Historic gains across the country. The papers are running "Starmer must resign" before the man has even had a bad week.
Here's my prediction, written on day 5 of this council term.
Reform doesn't lose the next general election to Labour, the Tories, or the Lib Dems. Reform loses it to itself.
Watch what's already happening. One councillor quit when he realised the job was unpaid. One stood down because he thought winning the seat made him an MP. One was elected over a year after he died. One was expelled four days in for alleged racist posts. 44 are already gone from last year's intake.
This isn't a vetting glitch. It's the operating system. Populist parties run on the promise that anyone can do politics, that the elites are faking the difficulty. Then they hand power to people who genuinely thought council work was a Westminster job, and the bluff collapses on contact with a parish meeting.
The cruelty of governing is that it's mostly boring. Drains. Bin contracts. Planning permission for a kebab shop. You can win an election on vibes. You cannot run a county council on them.
By 2029, the story won't be Reform's rise. It'll be the slow split between Farage's national brand and the hundreds of local councillors who quietly walked, got expelled, or simply stopped turning up. Labour or the Conservatives won't have to beat Reform. They'll just have to wait.
The thing that gets populist parties is never the opposition. It's the calendar.
Reform's biggest enemy isn't Labour or the Conservatives. It's the agenda for Tuesday's planning committee.
Ten years out, here's what I actually think happens.
Most people's daily lives look almost identical. They still commute, argue with their partner about the dishwasher, scroll something, sleep badly. The furniture of life barely moves.
But the invisible layer underneath gets weird. Most knowledge work is done with an AI sitting next to you, the way a calculator sits next to an accountant. Nobody calls it AI anymore. They just call it work.
That's the boring version. That's the one I'd bet on.
Now run the dial up 5x.
If AI moves five times faster than I expect, the thing that breaks first isn't the economy. It's meaning.
When a 19-year-old in their bedroom can build what used to take a 40-person company, the question stops being "what's possible" and becomes "what's worth doing." Most people have no answer to that question. We've outsourced it to our jobs for a hundred years.
You see the early signs already. Kids who can vibe-code a startup over a weekend but can't decide what to build. Designers shipping ten times more work and feeling ten times emptier. The bottleneck moves from skill to taste, and taste turns out to be the rarest thing in the room.
The countries that win the next decade won't be the ones with the best models. They'll be the ones whose people still know what they want.
And here's the part nobody's saying out loud. The accelerated version of AI doesn't make us obsolete. It makes us legible. It strips away every excuse we used to hide behind. "I didn't have time." "I didn't know how." "I couldn't afford it." All gone.
What's left is just you, your taste, and what you actually choose to do with a Tuesday.
AI isn't coming for your job. It's coming for your alibis.
Everyone’s screaming that AI will put intelligence into everything. Your fridge. Your fork. Your front door.
They’re wrong, and we already lived through this movie once.
Remember the Wi-Fi juicer? The smart salt shaker? The internet-connected toaster that needed a firmware update to make toast?
That was the “Internet of Things” revolution. It mostly produced landfill.
Adding intelligence to an object only matters if the object has a problem intelligence solves. My fridge doesn’t need to think. It needs to be cold. My washing machine doesn’t need an LLM. It needs to not break in seven years.
Here’s the bet I’d actually make. AI gets absurdly powerful in a small number of places. Software. Research. Anything that’s mostly thinking on a screen.
Meanwhile, most physical objects stay dumb. Because dumb is cheaper, more reliable, and doesn’t brick when the company’s servers shut down.
The revolution will be wildly uneven. Life-changing in some domains. Invisible in others. Annoying in the ones where AI gets shoved in to justify a $200 price hike on a coffee maker.
But here’s the part nobody’s pricing in. The opposite direction.
People are massively underestimating how much of AI’s value will be boring.
Not humanoid robots. Not AGI. Not your blender achieving consciousness.
Just: this thing that used to take 40 minutes now takes 4.
Compounded across a billion people doing slightly-less-tedious versions of their actual jobs, that’s the real revolution. Not cinematic. Not Twitter-thread-friendly. Probably bigger than the version everyone’s hyping.
The loudest predictions about AI are wrong in both directions at once.
Too cinematic about the things that won’t matter. Not boring enough about the things that will.
@IU_Wakilii I disagree with the idea of ‘don’t bring your money to Africa.’ Africa needs investment. Yes, we face challenges that require attention, but withholding investments isn’t the solution. Investment drives growth, creates opportunities, and addresses those very challenges.