What’s happening in the world of AI: Highlights The Invisible Shift: AI & The New Labour Equilibrium Full presentation discussion at [00:00] UK unemployment...
What’s happening in the world of AI: Highlights OpenClaw Just Acquired to OpenAI (And What It Means For Everyone) Peter Steinberger, the creator of...
Watch this video on our dedicated watch page
Better viewing experience with related videos and full-screen player
What’s happening in the world of AI:
No AI news worth covering, so instead I did a full walkthrough of something I get asked about constantly: how do you know if a business idea is any good before you waste months building it?
The answer is lean testing, and with AI, you can now do it in a single day. I ran through the full methodology live, showed my actual ads manager, my PostHog analytics, and a live A/B test I set up the previous day. Here's the complete framework.
Here's what most people do when starting their first business: they come up with an idea, obsess over it, polish it, build it in secret. The
y create business plans, hire people, spend money on logos and business cards and websites. They might spend six months, a year, even longer perfecting their product. Then they launch it.
Nobody cares. Nobody wanted what they built.
That's incredibly disheartening. And what happens next is usually worse. They go back into building mode, spend more time perfecting, still don't talk to anyone in their market, and end up trapped in this cycle until they quit and say "starting a business is too difficult." They attach their ego to the idea, and when the idea fails, they feel like they've failed as a person. Then they bounce off entrepreneurship entirely.
I see this pattern constantly. It doesn't have to be this way.
Instead of building the product and then launching, we're going to launch and then build.
We're going to sell something that doesn't yet exist. That sounds dodgy, but it's not. I’ll show you how we do this cleanly.
The process is: come up with an idea, build multiple sales pages for it, run traffic to those pages using Meta ads, and see whether people actually want it. Notice what's missing. We don't build the product. Not yet.
If enough people buy, we build it. If they don't, we've spent £100-200 (tops) and an afternoon instead of six months of our lives. Then we move on to the next idea.
Use AI to help you brainstorm, but start with something you're genuinely interested in and that solves a real pain. AI can give you infinite possibilities, so you need to narrow down. There's a free course on this at aiwithkyle.com called "Your First £1,000 AI Niche" that walks through the full idea generation process.
Critical rule: do not obsess over making it the “perfect” idea.
You can be broad at this stage because the testing process will refine the offer for you. Your idea just needs to solve a pain, have a clear outcome, and target a narrow audience. Don't go as broad as "how to lose weight." Something like "how to lose weight for men over 40 following a specific diet" is better.
As a live example: I'm currently testing a "7-Day AI Readiness Bootcamp" with another company. We haven't built the curriculum. Haven't written the content. We're selling the concept and using feedback to shape what the actual product will be. That's all you need at this point: the what, not the how.
Do not design logos. Do not hire people. Do not build the product. Not yet.
The sales page is more important than the product itself. If you can't get people through a sales page, it does not matter how good your product is. Nobody will ever see it.
I see so many people create genuinely excellent products but then can't write an offer around them. Nobody ever gets to the product because the sales page is rubbish. So you start with the sales page and use it to shape the product.
Build three or four variations using different hooks, headlines, images, and benefits. Use Lovable if you're not comfortable with vibe coding (you literally tell it "I want three variations of a sales page for [your idea]" and it creates and hosts them). If you are comfortable with coding tools, use Codex or Claude Code.
You need real strangers seeing your pages. If you already have an audience, you can test organically (I validated a £1,500 course by posting on TikTok, running a survey, then opening pre-sales before building anything, which led to about half a million in revenue). But if you don't have an audience, organic takes too long for a one-day test. That's where paid ads come in.
Specifically, Meta (Facebook+IG) ads.
Why Meta specifically? It's cheap and wide. LinkedIn ads are focused but very expensive. Google ads are problem-based (people searching for solutions), whereas Meta is interest-based. You can target broadly and let the algorithm figure out who responds. Meta also includes Instagram, so you get both platforms.
The budget: £20 per day for about a week. That's £140 total. You're not hiring a Meta ads consultant for thousands a month. You're spending beer money to find out if your business idea has legs. OK that’s quite a lot of beer but still you get the point!
For targeting: go broad. The only targeting I do is geographic (UK and US). No interest targeting, no demographic targeting. Why? Because the campaign itself will teach you who your customers are. My current campaign is showing me that the audience skewing towards 35+ males, nobody in 18-24, and heavy in the 50-60 bracket. That's incredibly valuable information. Older people have more money than younger people - that's good for my business! I'm learning this from the ads themselves rather than guessing.
For the ad creative: video works best. Don't use AI avatars, don't go faceless. Get on camera yourself. Take your sales page, feed it into AI, and ask for 10 video script variations that are congruent with the page. Then record them on your phone. The videos are not trying to sell anything. Their only job is to get people to click through to your sales page. Don't talk about price, don't explain the offer. Just generate enough curiosity to get the click.
I recently recorded 90 video variations (that's psychotic, don't do that!!). Do four or five. From my current test, one hook is absolutely destroying the others: "You are not late to artificial intelligence. If anything, you're early — more than 99% of people if you're watching this video right now." That one piece of information, which cost me ~£50 to discover, now feeds into my landing pages, my product, my entire ecosystem.
Just because someone signed up doesn't mean you have a viable business! Just because one person bought doesn't prove there's a large enough market. You need to set your metrics before the test starts. Otherwise you'll get emotionally attached and talk yourself into continuing something that isn't working.
Before you start, define your cost per click target, your cost per email collection (mine is under £2), your cost per purchase, and your desired margin given your product price. Then hold yourself to those numbers.
There's a story from Sony that illustrates why this matters. When they were developing the yellow Sports Walkman, they brought people into market research sessions. Everyone said they loved it. "It's so cool, so trendy, I can imagine using it in summer." Then Sony offered them the chance to buy one at a discounted price. Nobody bought it. People will tell you your idea is great. They will not necessarily open their wallets. Your mum will always say it's brilliant. Strangers with credit cards are the only signal that matters.
Oh yeah we love it! To buy though? No way.
This is also why taking payment is a much stronger signal than collecting emails. Couple of options here depending on your comfort level: a free email waitlist (weaker signal but lower risk), a small deposit like £5-10 (middle ground), or full pre-sale payment (strongest signal).
If you take payment and the idea doesn't get enough traction, you refund everyone, give them a bonus for their trouble, and keep them on your email list for when you launch something else. Be completely transparent about it.
After a week, one of two things happens.
It passes: You have sustainable cost per acquisition, consistent interest, and people are buying. Build the product you've pre-sold. Use the winning sales page to guide what you build. If the page focused on making money outperformed the ones about community and education, then your product should lean heavily into the money-making angle. The market has told you what it wants. Listen.
It fails: High cost per acquisition, low click-through, nobody buying. You've spent £150-200 and an afternoon. Kill it, move on. No sunk cost. No six months wasted. And crucially, you've skilled up through the process. Next time you'll be faster at building sales pages, more comfortable with Meta ads, and better at recording video. It's an investment, not a loss - you’ve learned a LOT.
If you're on the edge, you can extend the test another week. Refine the ads, refine the landing page. But I'd much rather you kill the idea and move on than cling to something mediocre. But don’t keep thinking you are “on the edge” just to justify what is a loser.
Your ideas aren't worth anything. I know that sounds harsh, but it's true. I meet people constantly who tell me about their brilliant business idea. It doesn't matter what you think about it. The only people who matter are customers, and whether they'll open their wallets. That's it.
The problem is that we tie our egos to our ideas. When the idea fails, we feel like we've failed personally. That's why speed matters. If you're testing an idea every day or two, there's no time to get attached. You throw one up, run some ads, it doesn't work, fine, next. Compare that to someone who spent 18 months building the perfect product with the perfect logo and launches to silence. That destroys people psychologically.
The iteration cycle is: idea → sales page → ads → data → kill or build. The whole thing can be done in a day once you've practised it. I set up my latest test (variations, PostHog tracking, Meta campaign) in an evening.
Oh and ne more thing: don't try to create something totally new. Don't try to make a market that doesn't exist. Apple didn't invent the MP3 player. They waited until Creative Labs and others had proven the market existed, then made the best version. There are two types of risk: competition risk and market risk. Market risk (no existing market) is far worse than competition risk. Competition means people are making money in that space. That's a good thing. Find what already works, find your niche within it, and test whether your specific angle resonates.
If you have a business idea sitting in your notes, get it out this weekend. Build a sales page, run some ads. By Monday you'll know if it's a goer.
Member Questions:
"What do you think of using Gumroad to sell books?"
Kyle's response: It's fine as a platform, but don't think that placing something on Gumroad means you'll automatically sell it. They have a Discover marketplace, but it doesn't really work. You still need to drive your own traffic. And they take 10%, which is steep.
If you're already sorting out your own distribution through ads or social media, you might as well set up Stripe and a website and sell directly. I've made about £20K on Gumroad over the years, but it's just a checkout tool, not a business strategy.
"Does Anthropic work well for creating a proper product now?"
Kyle's response: I use Claude Code for most things and it's excellent. That said, this week I've been using Codex 5.3, which now has a macOS app. It's the first time I've seen an OpenAI model beat Claude Code for my personal workflow. That was very exciting because I've been on Claude Code for nine months and nothing had come close until now. Both Opus 4.6 and Codex 5.3 are very good. The differentiation between frontier models isn't that large anymore, and they're all just going to keep getting better.
Make us a Preferred Source on Google and catch more of our coverage in your feeds.
What’s happening in the world of AI: Highlights "Something Big Is Happening" — The Essay Everyone Needs to Read Discussed at [00:00] An article...