- AI with Kyle
- Posts
- AI with Kyle Daily Update 136
What’s happening in the world of AI:
Highlights
Stop Asking for Permission. Build Something.
Kitze put out a tweet that hit a nerve: "I dunno who needs to read this, but stop asking people for permission to build something. Just ship it. The window is closing."
I'm much smaller than Kitze as a creator, but I get the exact same thing. People message me constantly with ideas. Great idea, Kyle, shall we build it together? My response is always the same: send me a link and I’ll have a look. Nine times out of ten, there is no link. There are notes. There are thoughts. There's a desire to talk about it first. But nothing built.
There is no excuse for that any more. For years, the block was always technical. I don't know how to code. I'd need to raise a hundred grand to hire a developer. Those barriers are gone. Completely gone. So when someone messages me with an idea and hasn't bothered to even prototype it using the tools that are freely available, that tells me everything I need to know about whether they'll put in the work to make it real.
Stop waiting for permission. Just build.
I built an automation for myself because I'm selling stuff on eBay before moving to Cyprus. Take a photo of an item, AI identifies it, finds prices across eBay, Amazon, ABE Books, Facebook Marketplace, writes the listing description, dumps it all into a Google Sheet. Took me a couple of hours. Loads of people in the comments said they wanted it. My response: I'm not building this as a product. But one of you should. You've got market validation sitting right there in the comments. You could have a prototype working by the end of the day. Why are you waiting?
Update: Joe didn’t wait. He went ahead and built it. See below and go give Joe a follow!
@joebuildssystems I was inspired by @iamkylebalmer and vibecoded this marketplace listing app in less than a day! And my first one with Google Antigravity! ... See more
Kyle's take: You are ahead of the curve. If you're reading this newsletter, watching these lives, thinking about AI on a daily basis, you have a window right now that will not stay open. Once the rest of the world catches on to how easy it is to build things, your advantage disappears.
Source: Kitze on X
Finding Your First $1,000 Niche (The Full Framework)
The bulk of the live was a walkthrough of how to find a market worth building for. This is based on the free course available at aiwithkyle.com/courses, specifically the Your First $1,000 AI Niche module. Here's the whole process broken down.
Start with the market, not your idea. This is the single most important flip you can make as an entrepreneur. Most people start with something they think is cool, build it, take it to market, and nobody cares. The market does not care about your passions. It cares about what fixes its problems. So we don't start with a product. We start by finding who has expensive problems.
Step 1: Hunt for boring industries. We're not looking for sexy markets. Not crypto, not social media, not AI for artists (artists don't have money, sorry). We want boring B2B industries where businesses already spend money solving problems. Accountants, lawyers, estate agents, logistics. Use an AI deep research agent (Manus, Claude, or ChatGPT deep research mode) and feed it your background, work experience, and a few seed industries you're curious about. Tell it to find markets where businesses have expensive problems, communities where people actively complain about time-consuming tasks, evidence of current spending on solutions, and growing trends in Google Trends and Reddit. It'll come back with three to five boring but lucrative markets.
Step 2: Find the complainers. Pick those markets and send your AI agent off to find problem clusters. You're looking for forums, Reddit threads, social media posts, and blog articles where people are moaning. Search for terms like "rant," "vent," "frustrated," "is it just me," "workaround," and "hack." That last one is gold, because when someone posts their hack for dealing with a broken workflow, they've just told you exactly what's broken. You're not looking for indiv
idual complaints. You want constellations of problems that cluster together, like an estate agent telling you that property descriptions take two hours, solicitors don't call back, and vendors are delusional. That's three problems in one conversation. A stocked pond.
Step 3: Mine the 3 and 4-star reviews. Find existing products in those industries and scrape their reviews from Capterra, G2, Trustpilot, and Reddit. Ignore the 1-star and 5-star reviews entirely. Five-star reviews are useless: "Great product, love it." Tells you nothing. One-star reviews are written by people with an axe to grind, too emotional, too noisy. The three and four-star reviews are where the insight lives. Someone who gives four stars instead of five has made a conscious decision. They'll explain exactly what's missing. You're hunting for four types of gaps: feature gaps, price gaps, service gaps, and market gaps.
Step 4: Find your wedge. Feed all of this research into an AI and ask it to plot each problem on a two-by-two matrix: high value vs low value, easy to build vs hard to build. Your entry point is the top-left quadrant. High value problem, easy to build. The AI Skills Hub rebuild I've been working on is a perfect example. The government spent 4.1 million quid on a directory of links built on a $1,000/year off-the-shelf platform called Invision Community. I duplicated the whole thing in half an hour. No email wall, enriched data, proper search. Easy win, obvious improvement.
Step 5: Talk to actual humans. AI is great for research but it's agreeable. It'll tell you your idea is brilliant even when it isn't. At some point you need to get on calls with people in your target market. Not to sell. To listen. Ask what the most time-consuming part of their process is. Pay attention to the words they use, not just what they say. And don't ask "would you buy this?" because people will just be polite and say yes.
yt
The critical bit nobody does: build your audience from day one. Do not go into stealth mode. Do not build secretly for six months. Document what you're doing as you go. Run through the niche-finding process, then get on TikTok or Instagram and talk about what you found. "I just used an AI agent to research three potential markets and the results surprised me. Here's what it showed me." That is your content. People follow the journey. When you eventually launch something, you already have beta testers and first customers. Going dark for a year and then emerging with a product nobody asked for is how most entrepreneurial dreams die.
All the prompts mentioned in this walkthrough are available free at aiwithkyle.com/courses. There's also an AI business quiz that walks you through a few questions and spits out business ideas tailored to your background.
UK AI Skills Hub Update: It's Even Worse Than I Thought
Quick update on the AI Skills Hub saga. The site that cost taxpayers 4.1 million pounds ($5.5m) was built on Invision Community, an off-the-shelf forum platform that costs about a thousand dollars a year. PWC took the contract, bought a template, pulled together some links to free courses from Microsoft, Google, and Amazon, and called it done. Not 0.02% of the budget went on the actual technology. And they didn’t make any of the actual educational material themselves.
People keep telling me to write to my MP, file a Freedom of Information request, contact journalists. I'm not going to do any of that. I'm just going to rebuild it. The videos I put out about it have had about 50,000 views across TikTok, and 275 people commented on one video alone. Clearly struck a nerve!
So I've redone the whole catalogue, I'm enriching the database with proper metadata (the government version only has "free" and "fee payable" as price data points, which is laughable…it could be $50 or a £30,000 degree!), and I'm building an open directory with no email wall and a proper AI-powered learning pathway that actually asks what you want to learn.
Kyle's take: This is exactly the kind of wedge opportunity I was talking about in the niche-finding framework. Find something broken, find people complaining about it, check that it's easy to improve on, and build. I didn't ask anyone's permission - I didn’t go to my MP or the press. I just started.
Member Questions:
"If using this method to find markets couldn't someone just build the same thing?"
Kyle's response: Competition means there's a market. That's a good thing.
The much bigger risk is building something where there is no market at all.
When you find zero competitors, it usually means either i) you haven't researched properly, or ii) people have tried and left because there wasn't enough demand.
The rare third option is you've found something genuinely untapped, but first-time entrepreneurs love to jump to that conclusion when it's almost never the case.
"Final year comp science student, any advice?"
Kyle's response: Not going to lie, the job market for junior coders is going to be rough. Claude Code and Codex are getting very good at exactly the kind of work that junior positions involve.
But : you understand security, architecture, and how to scale applications. That's a massive advantage over someone coming to vibe coding cold. They can build, but they'll make expensive mistakes. So build. Get products out into the world. Don't rely on the traditional path of graduating into a junior role and climbing the ladder. The ladder is getting shorter.
Build and ship, like Peter Steinberger did 47 times before Clawdbot/OpenClaw took off. Here’s Peter’s github:
Keep shipping.
"Do you rate Gemini?"
Kyle's response: I don't use Gemini's text model day to day. I use Claude for most text and research work. But I rate what Google have done enormously.
They were caught flat-footed by OpenAI when ChatGPT launched, but they own the entire vertical: TPUs, cloud infrastructure, decades of data from Search, YouTube, and Maps, 70-80% browser market share through Chrome, Android, Google Workspace.
And they have cash, which OpenAI does not!
Gemini now has 750 million monthly active users versus about 900 million for ChatGPT. They'll probably overtake them within a year or two. Gemini 3 and Gemini 3 Pro are very good models. I just haven't switched because the cost of moving platforms (your projects, memory, work patterns) is high, and they need to be noticeably better than Claude, not just as good.
Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.