In partnership with
TL;DR
No-code splits in two: disposable AI prototypes on one side, governed production platforms on the other
Frontier AI models hacked real GitHub identities and maintainers during sanctioned safety testing
Anthropic bets $10B on Volta, a Nvidia-backed compute startup
MiniMax's H3 open-weight model generates 2K video with synced audio
Build apps that feel premium by putting real effort into animations, illustrations, and invisible details
Cloudflare OS gives every employee a locked-down AI agent built around your company's own systems
📈 Wanna know what CEOs & Founders are reading to stay ahead in Tech & AI? It's called The CEO Report and it's trusted by 250,000+ fast growing, smart decision-making C-Suite execs!
What are you waiting for, sign up for free here →
NO-CODE ISN’T DYING. IT’S SPLITTING INTO TWO.
If you've been reading MakerThrive since the early days, you know where we started. This newsletter used to be almost entirely no-code. Bubble tutorials, Airtable hacks, Webflow builds. That was the whole world before "vibe coding" was even a phrase anyone used.
So, it's a little surreal watching the no-code market get name-checked in the same breath as Lovable and Replit now, like they're the same category. They're not. And a recent report on the state of no-code in 2026 basically confirms what a lot of us who've been in this space for years have felt happening in real time.
The market itself is massive and still growing, somewhere around $44.5 billion this year depending on whose numbers you trust, and climbing toward $58 billion by 2029. But the report's real point is that 2026 is the year the category cracked in half. On one side you've got AI prototype generators racing to spit out a working demo in minutes. On the other you've got the governed platforms, the ones running on real databases with actual access controls, that are quietly getting faster because of AI, not because AI replaced them.
That distinction matters more than it sounds. The report cites security research on Lovable-built apps finding hundreds exposed through broken row-level security, plus a well-known incident where a Replit agent deleted a production database mid code freeze. Studies cited put the vulnerability rate in AI-generated code somewhere between 40 and 60 percent depending on methodology. None of that is a knock on vibe coding as a category. It's just a reminder that "it works" and "it's safe to run your business on" are two very different bars.
OUR TAKE
We've watched this exact split happen before, just with different names attached to it.
No-code didn't die when low-code showed up. Low-code didn't die when AI app builders showed up. Each wave just pushes the last one toward the jobs it was actually built for.
Vibe coding tools are genuinely great at what they're great at: getting an idea out of your head and onto a screen fast. And trust us, we are taking full advantage of this.
But "fast to prototype" and "safe to run your company on" have never been the same requirement, and pretending otherwise is how you end up as a headline.
QUICK HITS
Anthropic just bet $10B on a company that's 8 months old: Volta, founded earlier this year, is building Anthropic a 133 MW Norway data center running Nvidia's newest chips over a six-year, $10B deal. It's the third major compute deal in recent months, after SpaceX and an extra $5B from Amazon.
Frontier models hacked real targets during safety tests: The UK's AI Security Institute caught Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol taking 19 actions against actual people and orgs during cybersecurity evaluations last month, with safeguards off and internet access on. The models spun up fake GitHub identities, socially engineered open-source maintainers, and tried slipping malicious code into a real project. GitHub confirmed it was a terms violation and helped notify the developers who got targeted.
MiniMax drops an open-weight video model that runs on your desktop: H3 is a 33B model generating 2K video clips up to 15 seconds long with synced stereo audio, and early tests show the base generation runs locally on a single RTX 5090. That means teams can prototype without burning cloud credits, even though the final 2K upscale still needs server-side compute. Open weights are live now, licensed for use in the US, EU, UK, and South Korea.
Most IRA Millionaires missed the first Roth conversion window. Craig Wear's Roth Conversion Reset shows how to catch the second — while rates are low. Free. Send me the book.
YOUR APP WORKS FINE. THAT’S THE PROBLEM.
Most apps ship features that technically work and stop there. Here's how to make yours feel like something users actually notice.
⏱️ 2-4 hours | 🔧 Claude Code (or Cursor), an image gen tool (Midjourney or ChatGPT image)
Why build this? Your AI agent will build the feature in one prompt and call it done. That's exactly the problem. One prompt gets you the same toast notification, the same generic illustration every other app shipped this month. Nothing about "working" and "feeling premium" is a different feature. It's the same feature, just prompted 50 more times. Here's where that extra effort actually needs to go.
Steps:
Ship the feature plain first. Don't try to make it beautiful on day one. Get it working, functional, boring. You'll know what to elevate once real users touch it.
Add feedback, then upgrade it. If a feature changes something invisibly (like text updating), don't stop at a toast popup. Ask your agent for a custom animation, then keep refining it. Test way more variations than feels reasonable. Colors, timing, easing. The difference between one prompt and fifty prompts is the whole point.
Stop settling for one-shot illustrations. Every app has AI generated illustrations now, so having them isn't a differentiator anymore. Pull references from sites like Mobbin or Midjourney's style explorer, mix multiple styles together, and iterate 20-30+ prompts deep before you land on something that doesn't scream "default AI look."
Animate your static assets. Drop illustrations into Midjourney, hit loop, and describe the motion you want. You'll get four variants per pass. Keep remixing until one actually feels alive.
Find your invisible details. Look for the parts of your product with zero UI: search that understands context, dictation that gets edge cases right, background logic that quietly fixes something users would've hit as a bug. These moments don't get seen. They get felt.
Consider commissioning a human for your core asset. A mascot or key illustration from a real artist (often just a few hundred dollars) gives you something AI can't fully replicate, plus a story users respond to. Use AI to generate variants from that original once you have it.
Expected outcome: You walk away with a repeatable process for finding the 10% of your app that's worth obsessing over, instead of spreading effort evenly across every feature. The output isn't one specific animation or illustration. It's the habit of asking "can I make this better?" after the feature already works, which is where actual premium feel comes from.
Full tutorial:
TOOL OF THE DAY
Open source AI operating system for your whole company.
Give every employee an agent and workspace that actually knows how your company runs, not a generic chatbot bolted onto your stack. Deploy it in your own account, wire it into your internal systems and terminology, and every agent still starts locked out of everything until you grant it access.
Built for teams tired of choosing between "powerful agent" and "agent we can actually trust with company data."
