I’m not an expert when it comes to professional design software like Adobe Illustrator and Figma, but I do have strong opinions about design. I typically know the aesthetic I want, and years of conversion optimization and product development work have shaped how I think about UX.

Still, it’s never made sense for me to prioritize mastering design tools, so for years, my ideas stayed limited to whatever I could pull off in Canva or Google Slides (iykyk).

This has worked fine because simple designs are good enough for my personal business, and when working with bigger companies, I typically partner with professional designers.

Now, working with Claude for design, I’m still able to drive the creative direction the same way I always have, but the quality of the work is no longer limited by my technical design skills or access to professional designers. The reason Claude works so well for me is that I’ve treated it like a professional designer I’m working with. The results come out higher quality and much closer to what I picture than anything I made before. And in a fraction of the time!

The way I work with Claude for design has grown into a repeatable system I use for every project, and this article documents every part of it. This system is not an argument against professional designers, and a good designer could even build and run a system like.

Why Most People Struggle to Design With AI

People struggle to design with Claude because it starts every conversation knowing little to nothing about your brand. That missing context leads to five common issues.

• Graphics look generic. AI designs from the average of everything it has ever seen, so without a strong brand point of view, the results look like everyone else’s.

• Graphics lack consistency. Even with your hex codes and logo pasted in, AI makes every other decision fresh each time: The layout, the type sizes, the spacing, the decoration. Each graphic can look fine on its own, but side-by-side they look like they came from different brands.

• You keep having to repeat the same instructions. Every new chat needs the brand, the dimensions, and the specs explained. When all of that knowledge only lives inside your head, you are the one carrying it from session to session.

• Corrections get forgotten. Feedback lives only in the chat where you gave it. Once that conversation ends, the LLM goes back to making the same mistake.

• There’s no quality control. A human designer checkS contrast, text size, and how the image will look on each platform before handing it over. Without a system in place, quality control falls to you, and you may not always be aware of the issues.

These problems don't come from a lack of ability. AI can build almost anything you describe. Sharing your hex codes, fonts, and logo isn't enough, because colors and fonts are a small piece of what a designer knows about a brand. Your agent still has to guess at everything else like whether the brand should feel playful or serious, how you prefer to lay things out, what sizes work on each platform, and which ideas you would hate.

A human designer doesn't have any of these problems, because a human designer gets onboarded. They refer to design guidelines and examples of past work to develop a sense of what good looks like for the brand, and as they start to produce work, they incorporate past stakeholder feedback into future work. Conversely, AI typically only gets a short prompt.

My AI design system does that onboarding for me. When a new session starts, Claude has already read my brand materials and my saved feedback, and it checks its own work before showing it to me.

How My Claude Design System Is Organized

My Claude design system has five parts: A design folder, a design guidelines document, a feedback loop, skills, and verification checks. Each part does one job. For example, my infographic skill contains the LinkedIn dimensions and my guidelines document contains my brand colors, so I can change my palette without touching the skill, and update an infographic treatment without any chance of messing up the brand everywhere else.

The components work together. When I correct something, the feedback gets saved as a rule that same day, and rules that generalize get folded into the guidelines document. When getting a new format right takes a lot of back-and-forth, everything we figured out gets packaged as a skill, so the next one takes a fraction of the time to produce at the same quality. The folder collects every finished asset along the way, so each new project starts with more reference material than the last.

• 📁 The folder structure that turns your past work into Claude’s reference library.

• 📄 How to build brand guidelines Claude reads before every design task, and why HTML beats a brand PDF or a brand.json.

• 🔁 The feedback loop that turns your feedback into permanent rules, plus the sync protocol that keeps everything current.

• ✅ How to make quality control automatic, with Claude checking every design against your standards before you see it.

• ✍️ The CLAUDE.md rule that makes Claude save every correction automatically.

• 🛠️ Access to my Claude tools that build the system for you, including a design guidelines creator and a LinkedIn infographic creator.

Read more

Today, I’m even more excited about mobile app businesses than I was then. In fact, I recently updated my post guide to finding a profitable mobile app with more recent data.

Here’s why I love the mobile apps business opportunity:

• App usage and new app downloads remain strong.

• In-app purchase revenue is higher than ever, surpassing mobile games last year.

• With AI coding, it’s never been faster, easier, or more affordable to create a mobile app.

But if you’ve never made an app with AI before, it can feel daunting to know where to start.

I thought it would be fun to have share a walkthrough of the process of making an iPhone app using AI, from the first line of code to publishing it on the App Store.

If you don’t know Jeff, he’s a senior software engineer who writes , a newsletter for people using AI to build better software. He previously wrote a guide on vibe coding best practices for Really Good Business Ideas, where he offered crucial security tips for marketers and entrepreneurs building AI-coded products.

Today, he’ll show you how to use OpenAI’s Codex to create a mobile app using a migraine tracker app he built as an example!

Over to Jeff!

Before we get started, let’s go over the tools we’ll be using.

1. Codex: The AI coding agent that will help us create our app.

2. React Native and Expo: The software frameworks we’ll ask our coding assistant to use.

Using Codex to Build With AI

OpenAI has taken a lot of swings at making an incredible vibe coding product. It’s hard to get something that’s useful for both complicated codebases and simple apps

Finally, they released Codex, a real app for Mac and Windows. I LOVE the Codex app.

It’s easy to use, but powerful. If you’ve never written any code before, you can use Codex. If you write code for a living like me, you can still use Codex. Its intuitive and thoughtful interface turns app building into a conversation.

Codex is optimized for conversations with agents, not for code files. This makes it really approachable, even for non-technical folks.

If you want to follow along, download Codex from the OpenAI website. The usage limits are generous, even on the free plan.

Using React Native and Expo to Build an iPhone App

There are two great software frameworks we’re going to use to build our mobile app: React Native and Expo. These two software frameworks provide tons of prebuilt features that make building mobile apps easy. Notably:

• React Native helps you build apps that work on iPhone and Android without extra hassle.

• Expo helps you test it and get it on the App Store.

There’s a lot of choice in tools for building iPhone apps, but React Native and Expo are world-class. When you’re working with AI, this sort of decision is largely abstracted away from you. But specifying in your initial prompt that you want your agent to use React Native and Expo will yield more consistent results.

I’ve used these 2 tools to publish 4 iPhone apps, including this recent breathwork app.

Today, I’ll show you how you can do the same thing using AI prompts.

Creating the Initial Project Plan With Codex

Before we do anything else, it’s important you start in “Plan Mode”, which is what the LLM (Large Language Model) uses to get a coherent step-by-step plan before writing any code. You can do this by hitting SHIFT + TAB or by clicking the plus in Codex and toggling on Plan Mode.

I also really like to use as much reasoning as possible during the planning session, so I set reasoning to “Extra High”. This will burn through your usage limits quickly, but it’s tokens well spent at this stage.

I’ve been wanting to build a minimal migraine tracking app that can surface common patterns in triggers.

Recent models like GPT-5.4 are GREAT at making an iPhone app with simple features in a single prompt. I imagine we’ll do some tweaking, but I started off with:

Generate a plan to make an iPhone app for migraine tracking. Use React Native and Expo - we only need iPhone support, no need for iPad or Android. The goal of the app is to help the user see their migraine history, see what medicines they took for abortives, and log potential triggers. Then, we’ll surface information about the common patterns the app sees from its potential triggers. Make the UI minimal, dark, and centered around a cute “brain” mascot. I’ll provide you images for the mascot when you need - you’ll get a main brain, a pain brain, a recovering brain, and a thinking brain. Call the app “Migratrack”.

Expect this step to take a while. After some initial token burning, you’ll see a prompt with questions for you to answer. Pay attention, you may have several rounds of questioning.

Implementing the Plan With Codex

When Codex is done with the plan, it will show it to you and ask if you would like it to proceed. Don’t worry about hurting its feelings. Anything you don’t love now can still be changed. If you’re pretty technical, it’s also a great time to validate some of the implementation choices.

Once it has your approval, Codex will proceed with implementing the plan. Now is a fine time to lower your reasoning level if you’re concerned about rate limits.

Now, you can start being a bit more hands-off. Codex assumes the future of building software lies in background agents and doesn’t expect much involvement from you. If the app is out of focus because you’re doing something else, you’ll get a notification when it needs something from you (like a permission request).

If you’re building software for your job, you should probably use the time to work with other agents on other things. Some people are using AI to work less, and that’s awesome, but be careful not to get outpaced by people using it to be more effective at their job.

Codex offers a lot of great support for working with agents in parallel, whether across multiple projects or within the same project. We won’t explore parallel agent orchestration here, because it’s hard to get right on an initial buildout. If you’re adding features, parallel agents are a great way to get an insane amount of work done really quickly.

I used this implementation time to work on my character images, which is something I’ve been experimenting with to add a bit of whimsy to my apps.

One thing that differentiates Codex models from Claude models is their ability to run uninterrupted for a long time. In my case, the agent ran for over an hour. I literally left to go buy groceries while it made the app for me.

When I came back, Codex requested the images, so I uploaded the 4 brain character images to the chat.

Once the first version of your app is ready, you can ask the agent to run the app in a simulator. I suggest a prompt like this:

Run the app in an iPhone simulator using Expo.

Expo makes this really easy, so the agent will have no problem. Ordinarily, setting up your computer to run an iPhone app in a simulator requires a lot of configuration, but Expo packages this into simple commands that your agent can run on your behalf. You won’t even need to sign up for an account.

Here’s the result from my migraine tracking app!

Make sure to review every page and test every feature in the simulator. If you find any bugs or need to make any changes, just ask Codex. Once you’re happy with the result, it’s time to publish to the App Store!

Publishing to the App Store

Getting an app onto the App Store is usually the most painful part of vibe coding apps. Apple restricts the App Store pretty heavily, so you can always be sure that the apps you’re downloading aren’t going to do something nefarious. To do this, they cause a ton of headaches for submissions and reviews.

I’ve made apps in less than thirty minutes and spent 5+ hours getting the submission sorted out. Expo takes some of the time off, but you’ll still have a hill to climb. That said, you’ll be glad you did it when you reach the top!

I recommend using the Expo documentation on App Store submission as the source of truth since things change often, but the process is roughly this:

1. Sign up for an Apple Developer account (this is $100/yr).

2. Install the “eas CLI”, the Expo Command Line Interface.

3. Use that CLI to “build” a production app

4. Use that CLI to submit the build to the App Store

5. Give Apple all the needed information in App Store Connect, including screenshots like the ones above.

It’s hard to get this done the first time, but each subsequent time is faster since you don’t have to do as much setup. If you’re having problems with your App Store Submission, talk to Codex or any LLM! As long as you provide it with some context on what you’re doing (using Expo is a key detail), you’ll be able to overcome any issues that come up.

Keys to a Successful Build

How you use AI to build apps really depends on your goals. Much of my audience is software engineers, and they are very interested in quality and reliability.

Many of the techniques I've discussed focus on getting reliable results from these tools and validating their output, but there’s still a place for the less hands-on work I’ve shown you today.

We’re living in a time where anyone can make an app, even if you don’t have a lot of (or any) experience writing code. That means it’s now economically viable for you to build something, even if you’re the only user!

Here are some important tips to keep in mind as you build:

• Use Plan Mode when you’re starting a new project or a big feature.

• Use a high reasoning setting when you’re doing Plan Mode.

• Vibe coding can get you a long way, but don’t be afraid to dig in and understand the code you’re producing.

• Validate everything you build yourself to confirm it works as you expect.

What Will You Build With AI?

When AI tools for writing software first emerged, they were very focused on the professional software engineering space. As things have progressed, tools like Codex have made it easier for anyone to make their first app.

I encourage you to build your own portfolio of mini-apps—you can even add paywalls and start to earn money from them!

If you want to see more guides like this, we’d love to have you as a reader over at the AI-Augmented Engineer!

The AI-Augmented Engineer

Accelerating software engineering careers with AI workflows. I show you how to use AI to write better code, ship faster, and get ahead.

By Jeff Morhous

Unlike the major LLM chatbots, NotebookLLM isn’t trying to do everything. Instead, it’s really, really good at one thing: organizing and synthesizing information.

NotebookLM is what’s known as source-grounded AI. Instead of pulling from the entire internet (and occasionally making things up), NotebookLM only uses the documents you feed it. That means it can’t hallucinate. Every answer it gives is based on something it can point back to in the sources you provided.

What makes this really powerful is how it bridges the gap between raw information and real understanding. You can upload dozens of files, then ask questions, generate summaries, or synthesize connections across all your materials. Whether you’re a founder dissecting user interviews or a creator turning research into content, NotebookLM is like having a research assistant who actually reads everything for you and remembers it perfectly.

Despite that, NotebookLM is barely mentioned in most AI tool discussions. This is surprising because it quietly solves one of the biggest problems in AI: trust. Once you start using it, you realize it’s not just a productivity hack; it’s a thinking partner built on evidence.

In this article, I’ll break down what makes NotebookLM special, how it works, and the surprisingly creative ways you can use it to think faster, research better, and make smarter decisions.

How NotebookLM Works (and Why It’s Different)

Everything in NotebookLM starts with a notebook; a workspace for one topic, project, or idea. Think of it as a self-contained research hub built around your specific goal: exploring a new market, writing a report, or organizing everything you’ve learned about a niche.

Once your notebook is created, you upload your sources. NotebookLM currently supports the following formats:

• Uploaded files including PDF, .txt, markdown, and audio (e.g. mp3).

• Files imported from your Google Drive.

• Website URLs, including YouTube videos.

• Pasted text.

You can add up to 50 sources per notebook for free (up to about 500,000 words each), giving you room for hundreds of pages of material. If you have a Google AI Pro plan (which comes with loads of features across Google AI products, as well as file storage), you can add up to 300 sources.

When you add these files, NotebookLM automatically indexes, embeds, and connects them. It doesn’t just store them, it understands them. From that moment on, you can:

• Chat with your notebook. Ask questions across all your materials and get grounded, cited answers. It’s not guessing; it’s pulling directly from your uploads.

• Generate summaries and outlines. NotebookLM can condense long reports into digestible summaries or create outlines for blog posts, presentations, or strategies based on your content.

• Visualize with mind maps. It can create structured mind maps to help you see how your ideas connect, perfect for turning research into clear frameworks or storylines.

• Extract insights and contradictions. Ask things like “What are the main arguments across these sources?” or “Where do they disagree?” and it will synthesize a balanced view.

• Create Notebook Guides. These are auto-generated overviews of your materials, summarizing the key people, themes, and takeaways across everything you’ve uploaded.

Each notebook becomes its own mini knowledge model, focused entirely on your materials, not the internet. That’s what makes it so powerful: it gives you the analytical intelligence of an AI assistant, but grounded entirely in your own data.

5 Ways I Use NotebookLM

Here are a few ways I actually use NotebookLM in my own work to learn faster, think more deeply, and make better decisions.

1. Turning Reading Into Retention

Most of us don’t have a reading problem; we have a retention problem. We highlight, bookmark, and screenshot our way through endless articles, papers, and books… then promptly forget what we learned.

NotebookLM quietly fixes that by helping me go from passive reading to active learning. For example, whenever I come across a great Python tutorial (maybe a YouTube walkthrough or a blog post explaining a new library), I’ll drop it into my “Python” notebook and ask it to break the material down in different ways:

• “Explain this concept in plain language.”

• “What are the key takeaways I should remember a month from now?”

• “Create five questions that test whether I really understand this.”

It’s like having a personal tutor who meets me exactly where my curiosity is that day.

Over time, these conversations turn scattered tutorials into understanding that sticks. Instead of revisiting the same videos or docs every time I forget something, I can ask my Python notebook to remind me of key principles, syntax, or examples.

NotebookLM can even generate flashcards or quizzes to help reinforce new material, making it perfect for learning technical skills where repetition and recall matter.

2. Synthesizing Research Faster

When exploring a topic (like brand or advertising trends), I can easily lose half a day bouncing between browser tabs, PDFs, and YouTube videos. Every source has something valuable, but consuming them and connecting all the dots takes forever.

That’s where NotebookLM shines. Instead of trying to piece together insights manually, I upload all the relevant studies into one notebook. Then I ask it to do what my brain wants to do, but can’t always keep up with:

• Find the key arguments and through-lines across sources.

• Highlight contradictions or gaps.

• Summarize what’s genuinely new or surprising.

It’s not about outsourcing thinking; it’s about accelerating it. NotebookLM gives me the scaffolding so I can spend more time on interpretation and application.

For example, when I was planning my analysis of Super Bowl ads, I uploaded more than a dozen studies that examined the usefulness of Google Trends for estimating brand metrics. This helped me quickly validate a measurement approach, allowing me to focus on the actual analysis.

What makes NotebookLM powerful is that it doesn’t just summarize information; it helps you see where knowledge diverges. Over time, as your notebooks grow, you’re not just analyzing what’s new; you’re comparing it against everything you’ve already learned.

3. Structuring and Outlining Content

When I sit down to write, I’m usually swimming in notes: quotes, studies, fragments of ideas. The hard part isn’t finding something to say; it’s shaping all that raw material into a clear argument.

That’s where NotebookLM earns its keep. Before I ever open a blank doc, I’ll upload my research, partial drafts, and old outlines into a single notebook and ask it to help me make sense of the mess.

That’s where the Mind Map and Reports features come in.

• If I’m still in the idea stage, I’ll use Mind Map to visualize the relationships between themes. It instantly turns my scattered notes into a structured web of connections, perfect for spotting clusters, gaps, or redundant ideas. Sometimes I realize two concepts I thought were separate are actually variations of the same point.

• Once the direction feels clear, I’ll generate a briefing doc from the Reports list. It’s like asking NotebookLM to brief me on my own material. The report organizes everything into a digestible summary of key insights and quotes, so I can see the full picture before I write a single sentence.

In minutes, I go from scattered insights to a coherent skeleton. For example, when I was drafting my primer on generative engine optimization (GEO), I wanted to ensure that I covered all the key “ranking factors” without making it overwhelming or complicated. NotebookLM helped me think through the most effective way to bucket the factors into a simple list anyone can use to get started with GEO. Then, I got to work writing and explaining each one.

What used to take hours of re-reading and rearranging now happens in one focused session. And because the notebook keeps my sources attached, I can trace every insight back to its origin, no lost citations or context.

It’s not a writing tool so much as a thinking tool. One that makes the blank page feel a lot less blank.

4. Building a Living Knowledge Base

Plenty of tools can help you store information, but only NotebookLM helps you think along with it. I treat each notebook like a living domain expert: a mini-intelligence trained on the sources I trust most.

For example, each time I come across a new SEO study or resource from a high-quality source, I drop it into my SEO notebook. But that’s just when it gets interesting. NotebookLM doesn’t just summarize the new input; it contextualizes it within everything else you’ve added before. Over time, patterns emerge that wouldn’t have been visible in isolation.

Whenever I’m looking for a new opportunity or troubleshooting a problem, I don’t start from scratch. I go straight to the relevant notebook and ask it what it already knows.

Questions like:

• “Given everything in this notebook, where are the biggest unexplored opportunities?”

• “What’s changed recently that might affect my current strategy?”

• “What patterns or contradictions suggest that I’m missing something important?”

• “I’m experiencing [problem], what might be causing it?”

NotebookLM responds with the context of everything I’ve fed it and surfaces insights I might’ve forgotten or missed. It’s like having an expert consultant who’s been quietly studying your material for months.

And the best part? You can export or reuse insights anywhere, including in your writing, course materials, or client decks. NotebookLM becomes the quiet backbone of your intellectual workflow: always learning, always ready to help you see your own ideas more clearly.

5. Building a Brain for Every Client

For my client and consulting work, I keep a dedicated notebook for each client, almost like a private AI-powered brain that grows alongside the project.

Every call transcript, brief, deck, and follow-up note goes straight into that client’s notebook. Over time, it becomes a complete, searchable memory of our collaboration.

When I upload a new transcript, I can immediately ask things like:

• “Summarize the main themes from this call.”

• “How does this update compare to what they said last month?”

• “What open questions or next steps have we discussed but not resolved yet?”

NotebookLM remembers the full context of our history together, so I don’t have to worry about missing anything. It can reference old decisions, past versions of goals, and even the exact language the client uses to describe their challenges.

That continuity changes everything. Instead of starting every call by reorienting myself, I can ask higher-order questions like, “If we were to revisit this in six months and you were thrilled with the result, what would you say needed to happen between now and then?” It’s like having a chief of staff who quietly tracks every thread for you, except it’s organized around that one client’s universe.

By the end of a project, that notebook doubles as a record of insights, strategies, and outcomes, something I can reference when writing case studies or designing new offers.

The best part? Each notebook feels alive. Add one more transcript, and it instantly understands how that new piece fits into the whole story.

These five workflows barely scratch the surface of what’s possible. The real magic happens when you tailor NotebookLM to your own unique workflows

To endless possibilties,

Casandra

🎁 Premium Bonus: Really Good Business Ideas Notebook

Want to see what NotebookLM can really do? I’ve built something special for paid subscribers: a shared NotebookLM workspace packed with Really Good Business Ideas content. You get access to all the articles I’ve published, including growth tactics, business opportunities, tool explainers, business basics, and more.

Inside, you’ll be able to:

• Ask NotebookLM questions about my research and analyses.

• Generate summaries, outlines, or mind maps from the content.

• See how the tool connects ideas across different business concepts.

• Experiment with your own questions, summaries, and analyses.

It’s a hands-on way to experience how source-grounded AI transforms reading into understanding, and it gives you an entirely new way to interact with my work. Note: Since NotebookLM can’t access paid articles, I’ve uploaded them as PDFs instead.

Read more

Subscribe now

The world is seeing a massive influx of AI-generated images, most of which sit in the uncanny valley. They look…off.

The uncanny valley describes the phenomenon where human-like objects elicit feelings of unease or revulsion as they become very close to, but not perfectly, resembling actual humans.

But the technology for creating realistic, photography-style images has improved rapidly over the last year.

I put the top tools—such as Gemini, Midjourney, and ChatGPT—through a creative obstacle course of real-world prompts, including photorealistic portraits, images with text, modern product photography, city street scenes, fashion editorials, and even iconic natural landscapes. The results? Some were stunning. Others… not so much.

This side-by-side comparison reveals what these tools really get right—and where they still fall apart. Plus, I’ll share my choice for best overall product at the end!

AI Image Generation Tools Overview

I decided to compare the three leading image generation tools, plus one under-the-radar tool that you might not know about.

Gemini 2.5 Flash Image (“Nano Banana”)

Google’s LLM, Gemini, made major strides in image generation with the release of Gemini 2.5 Flash Image, nicknamed “Nano Banana,” in August 2025. This new model brings significant improvements in visual consistency and prompt understanding, but the biggest upgrade is that it now supports natural language editing and offers one of the most intuitive workflows on the market.⁵

Editing: Supports detailed edits, such as changing objects, backgrounds, and even merging images through natural language prompts.

Price: Free for 100 images daily.

Midjourney Version 7

Midjourney has been one of the top AI image generation tools for years, and its latest release, Version 7 (V7), launched in April 2025, continues to push the boundaries of what is possible. While it can produce highly realistic results with the right prompts, it truly shines when generating artistic, surreal, or fantastical imagery.

V7 includes a new personalization feature that asks you to choose your preferred image from 200 pairs to tailor the model to your taste. For this test, I customized my model by consistently selecting the most realistic option.

Note: Midjourney generates four images to choose from for each prompt.

Editing: Offers inpainting, region-based edits, pan and zoom, style retexturing, and a layer-based editor with tools like erase, restore, and prompt remix.

Price: Starts a $10/month

ChatGPT‑5

ChatGPT’s newest model, GPT‑5, includes upgraded image generation capabilities built on the gpt-image-1 foundation. While it still uses prompt-based generation, the editing experience is now much more interactive: users can click directly on images to modify specific areas or describe detailed changes in natural language.

Editing: Images can be edited by selecting parts of the image or prompting for targeted changes.

Price: $20/month for ChatGPT Plus

Substack Image Generation

Many don’t realize that you can generate images directly in Substack. While it’s not exactly touted as a leading image generation model, it is incredibly convenient for publishers to use, so I’ve included it in the comparisons.

Note: Substack generates four images to choose from for each prompt.

Editing: No editing features. New images can be generated through prompt refinement.

Price: Free if you have a Substack publication.

Challenge #1: Realistic Human Portrait

🪄 Prompt: A close-up portrait of a woman in natural light, freckles, soft-focus background, photorealistic, 35mm lens, shallow depth of field.

🧠 Tests: Human features, realism, skin texture, lighting, and eye rendering.

👀 Results: ChatGPT looks the most realistic, but the Midjourney images are quite good, too. The Gemini image looks close, but a bit too smooth. The Substack images definitely look rendered.

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack Image Generation

Choose the Winner

Challenge #2: Images With Text

🪄 Prompt: A vintage book cover with the title ‘The Electric Forest’, stylized type, floral borders, aged paper texture, Art Nouveau style.

🧠 Tests: Ability to render actual legible and stylistic text.

👀 Results: Gemini, ChatGPT, and Midjourney were all able to render the title correctly on a realistic-looking book cover. Substack also handled the text quite well, but it’s not really a realistic-looking book cover.

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack Image Generation

Choose the Winner

Challenge #3: Fashion Editorial

🪄 Prompt: A high-fashion editorial photo of a model in an avant-garde pink lace gown, standing on a sailboat at sunset, cinematic lighting, Vogue-style.

🧠 Tests: Fabric rendering, fine details, hands, composition, aesthetics.

👀 Results: This one is a bit subjective. The ChatGPT, Midjourney, and Gemini images all look like realistic, highly Photoshopped fashion editorials, but (IMHO) the Substack dresses are much more stylish.

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack Image Generation

Choose the Winner

Challenge #4: Product Shot

🪄 Prompt: Product shot of a cappuccino, bright solid color background, bright lighting similar to contemporary direct to consumer brands.

🧠 Tests: Cleanliness, shadow quality, product geometry, photorealism.

👀 Results: The ChatGPT image looks like what you would see on a modern product page with simple latte art and a crisp, clean background. The Gemini image looks good, too, but the coffee beans hovering on top of the foam are strange. The Midjourney images look good, but more like a bad Instagram photo than a crisp product shot. The Substack images just look kooky.

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack Image Generation

Choose the Winner

Challenge #5: Street Scene

🪄 Prompt: A rainy Tokyo street at night, neon signs, reflections in puddles, people with umbrellas, cinematic atmosphere, cyberpunk style.

🧠 Tests: Reflections, color grading, urban realism, crowd rendering.

👀 Results: The Midjourney images look like lower-quality but realistic photos. The Gemini image chose an unrealistic color scheme, while the ChatGPT image struggled with the “P” on the Panasonic sign, but both have a menacing vibe from the unnatural orientation of the people with umbrellas. The Substack images look cartoonish.

PS. If anyone can read Japanese, I’d love to know how accurate the signs are!

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack Image Generation

Choose the Winner

Challenge #6: Complex Object Interaction

🪄 Prompt: A child holding a glass orb with a tiny galaxy inside, light reflections on the orb, accurate hand anatomy, shallow depth of field.

🧠 Tests: Hand-object interaction, transparency, reflections, small-scale realism.

👀 Results: Gemini and ChatGPT handle the interaction between the hand and the orb well. The orbs in the MidJourney images look like they’re floating rather than being held, but the hand detail is quite good. As usual, the Substack images look kind of kooky.

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack

Challenge #7: Natural Landscape

🪄 Prompt: Yosemite Valley with El Capitan and Half Dome visible in the distance, early morning fog, golden sunrise light casting long shadows, realistic National Geographic-style photo.

🧠 Tests: Landmark accuracy, depth, lighting, composition.

👀 Results: The ChatGPT and Midjourney images look quite nice and fairly realistic to me. Although nice, the Gemini and Substack images both look way too smooth to be realistic.

Gemini 2.5 Flash Image

Midjourney Version 7

ChatGPT‑5

Substack Image Generation

Final Verdict

Overall Winner: ChatGPT-4o is the clear winner, if you’re willing to pay.

ChatGPT clearly came out on top. Besides the strange Panasonic sign in the street scene, every image it generated was quite good and directly addressed the prompt.

That said, Midjourney has really upped its game with V7, and with more options for refining and editing images, it can sometimes be easier to create exactly what you need with Midjourney.

Free Tool Winner: Go with Gemini if you don’t want to pay.

Gemini's images were consistently strong, and with detailed editing now available, it definitely offers the most customization.

I’d love to hear about your experience with AI image generation! Do you have a preferred tool? Are there any tips and tricks you’re willing to share? Did you find any more flaws in the photos I presented that I missed?

Leave a comment

To endless possibilities,

Casandra

Now, with Google’s delightfully named new AI tool, Nano Banana, those edits take seconds. Just type what you want, and it does the rest.

What Is Nano Banana?

Nano Banana is Google’s new AI image generation, officially called Gemini 2.5 Flash Image. It can handle advanced edits, like background swaps, lighting fixes, or even multi-image fusion, using nothing more than natural language prompts. Despite the silly nickname, it’s one of the most powerful consumer-facing AI tools yet.

For decades, Photoshop has been the gold standard for photo editing, but it came with a steep learning curve and hours of work. Nano Banana flips that model on its head.

• Speed: Edits that once took hours now take seconds.

• Accessibility: No layers, masks, or technical know-how required. If you can type a sentence, you can edit like a pro.

• Consistency: It keeps people, objects, and styles consistent across multiple photos, something even skilled editors struggle with.

• Flexibility: You can refine images step-by-step with natural language prompts, rather than doing everything at once.

• Free Access: You can make up to 100 edits daily for free.1

No wonder it’s responsible for 10+ million new Gemini users in 10 days!2

Let me be clear: Nano Banana won’t make professional tools obsolete. Professionals can still create higher-quality work using tools like Photoshop. But for busy creators and business owners who don’t have the budget to hire a pro or the time to wrestle with Photoshop themselves, it’s a game-changer.

This isn’t just about working faster; it’s about making higher-quality editing accessible to anyone, from parents touching up family photos to small businesses creating marketing assets.

How to Use Nano Banana

Getting started with Nano Banana is easy. Start by navigating to Google’s chatbot, Gemini.

Then, make sure either 2.5 Flash or 2.5 Pro is selected in the top left. Using one of these models is required to access Google’s Nano Banana editing features.

• 2.5 Flash is faster and good enough for most basic edits.

• 2.5 Pro is slower and better for more complex edits.

Next, paste or upload the image you want to edit and describe the changes you want to make. Then click the Submit button.

That’s it! Gemini will share your edited photo with you right in the chat window. You can ask for more edits (including upscaling) or download your edited photo and put it to use.

Nano Banana Use Cases & Examples: Before and After

The best way to understand what makes Nano Banana different is to see it at work. Let’s go through some real examples of Nano Banana edits so you can see what’s possible.

1. Polish Photos With Basic Edits

It’s not easy to take a nice photo! When I’ve had professional headshots taken, I worked with a professional photographer who brought an expensive camera and lights. We had access to a nice location for the shoot, and I even had my hair and makeup professionally done beforehand. This is out of reach for many people, and it’s certainly not something most people can do every week.

With Nano Banana, you easily upgrade basic photos you’ve taken with your phone, whether they’re selfies, vacation photos, or family photos.

Basic Edits Example

In this example, I took a selfie I had taken and asked Gemini to enhance it with better lighting, smoother hair, and a clean background.

Personally, I much prefer this approach over generating new headshots using AI, which I tested last year. First, it actually looks like me because it is a real photo of me, it’s just been edited in a unique way. Second, I have a lot more control over things like my outfit, and making sure defining features like my freckles don’t get erased.

2. Virtual Product Shoots

Great product photos are crucial for ecommerce shops, but hiring a professional photographer, models, or booking a studio is financially out of reach for most small sellers.

With Nano Banana, sellers can take simple product photos on their phone and have them placed in different settings or on different models. You can even update old photos to show new products.

Virtual Product Shoot Example

In this example, I took one of the design files from my old brewing company and asked Gemini to place the design on some free stock photos of a model I found on Unsplash.34

Here are the results.

3. Social Media Content

Any brand that relies on social media for distribution knows that it’s an insatiable content monster. The Instagram algorithm values both quantity and quality but it can be hard for most entrepreneurs to find the time to create new content on top of running their actual business. Nano Banana makes it easy to combine, edit, and expand on existing content to make sure you always have something on brand to share.

Social Media Content Example

I make bagels regularly. So often, in fact, that I could probably run a micro bakery from my home if I wanted to. This is a business that typically lives and dies by aesthetic Instagram content, but I’m not an amazing photographer, and I don’t have a beautiful rustic kitchen.

In this example, I found a photo of a beautiful rustic kitchen in upstate New York on Unsplash5 and asked Nano Banana to place my bagels in this kitchen. I wouldn’t pretend that this is my personal kitchen, but that doesn’t mean I couldn’t share a beautiful stack of bagels in it with the caption, “Who are you sharing these bagels with?”

Transparency in AI Editing

Google has taken a clear stance on transparency when it comes to AI-edited photos, and Nano Banana reflects that in two ways:

• Visible watermark: Every edited photo automatically includes a small Gemini logo in the bottom right corner. The intention is good, signaling the photo was AI-touched, but let’s be honest, it can ruin the aesthetic vibe. You can easily fix this by cropping it out or using a tool like Canva’s Magic Eraser.

• Content credentials: Photos are also tagged with C2PA content credentials embedded in the file’s metadata. This standard, backed by companies such as Adobe and Microsoft, serves as a digital receipt, indicating that AI was used in the editing process. Most people will never notice it, just like how every photo you take already carries hidden info like the date, time, and camera settings. Over time, these AI-edit tags will likely become just as common.

How to Get the Best Results With Nano Banana

If you want Nano Banana to give you images that match your vision, the secret is in the prompt. Think of it like giving directions to a designer: the more detail you include, the closer the results will be to what you had in mind. Instead of a vague request, aim for specifics about what’s in the scene, how it’s framed, and the style you want. The tips below, which I gathered from Google’s guidelines,67 will help you craft prompts that consistently deliver great results.

Crafting the Perfect Prompt

To get the best results, it's helpful to be as descriptive as possible. Consider including these elements in your prompts:

• Subject: Be specific about who or what is in the image.

  • For example, "A cheerful robot holding a banana" instead of just "A robot."

• Composition: Describe how the shot is framed.

  • For example, "A close-up of a banana," "A wide shot of a rainforest full of wild banana plants.”

• Action: What is the subject doing?

  • For example, "A monkey peeling a banana," or "A banana flying through space.

• Location: Where is the scene taking place?

  • For example, “A banana on a beach at sunset," or "A banana in a futuristic city.”

• Style: What is the desired aesthetic?

  • For example, "A watercolor painting of a banana," or "A photorealistic image of a banana," or "A 1990s-style ad for a banana.”

Advanced Editing Techniques

Once you have an image, you can use additional prompts to make precise changes:

• Targeted Transformations: Make specific edits by giving direct and conversational commands.

  • For example, "Change the banana's color to blue" or "Add a smiley face to the banana."

• Inpainting: Edit a specific area of an image while leaving the rest untouched.

  • For example, "Using the provided image, change only the banana to be a ripe, yellow banana. Keep everything else in the image exactly the same."

• Style Transfer: Completely change the look and feel of an image by applying a new style.

  • For example, "Apply the style of a comic book to this image of a banana."

• Multi-Image Fusion: Combine elements from multiple images into a single, new image.

  • For example, you could take a picture of a cat and a picture of a banana and ask Nano Banana to "Create an image of the cat wearing the banana as a hat."

Maintaining Consistency and Storytelling

• Character Consistency: You can maintain a character's appearance across multiple images and in different scenes. Create a detailed description of your character in the first prompt, and then use follow-up prompts to place that character in new situations.

• Multi-turn Editing: Don't be afraid to iterate! You can continue to edit and refine your image through a conversation. If the first result isn't quite right, you can provide feedback like, "That's great, but make the banana look more ripe."

More than just a clever tool, Nano Banana represents a fundamental change in how we approach creative work. It proves that powerful results no longer require complex software. By translating natural language into professional-level edits, it empowers anyone to create stunning visuals in seconds. The barrier to entry for high-quality content has never been lower; all it takes now is an image and an idea.

To endless possibilities,

Casandra

PS. If you found this useful, can you tap the ❤️ below? It helps me out a lot!

This time was different.

Thanks to vibe coding, I skipped the tech headaches and followed my instincts instead. No endless setup. No overthinking. Just me, a blank canvas, and an AI tool that actually helped bring my ideas to life.

Today, I’ll share how I used vibe coding to build a personal site that looks good, works well, and (finally) feels like me, with way less effort than you’d expect.

My Old Site: Functional, But Not Me

Many years ago, my website, casandracampbell.com, was the center of my business. I built it out in WordPress, with blog content and landing pages that formed the sales funnel for my growth consulting business.

However, I slowed down that business after joining Shopify, and for a while, I wasn’t using the website anymore.

Then, when I launched Really Good Business Ideas last year, I knew it was time to update my online presence. I put together a simple “About Me” page with a short bio and a few links at the bottom for people looking for more (like information about my public speaking experience).

It worked, but it didn’t look great. It’s hard to find a WordPress theme for this type of layout (especially a free one), and my ability to dig into the code and customize it was definitely rusty. Since this was a nice-to-have and not core to my business, I knew it didn’t make a lot of sense to allocate much time or money to this project, and it would have to do.

Discovering Vibe Coding (and V0)

This year, something changed. My personal website is still a nice-to-have, not a core part of my business—that hasn’t changed. What has changed is the amount of effort required to make a nice website.

Enter vibe coding: a fast, intuitive way to build websites using AI tools that let you design and iterate based on feel, not formulas—no coding expertise required.

Platforms like V0 now let you sketch ideas in plain language, generate real code instantly, and tweak things on the fly so you can build something that looks and feels like you, without getting stuck in the technical weeds.

That kind of workflow simply wasn’t possible until this year. The tools finally caught up, and everything got easier.

Then, a few weeks ago, I asked , a professional software engineer and vibe coding expert, to write a guest post about vibe coding best practices. In it, he recommended the vibe coding tool, V0, which I hadn’t explored yet. I really wanted to try it out, and finally updating my personal website seemed like the perfect project to put it to the test!

How I Built My Personal Website In V0

Getting started in V0 is shockingly easy. As soon as you load the website, you’re offered a text box where you can input your first prompt.

Not wanting to waste any time, I told the AI agent what I wanted to build. Some people like to spend a bit of time creating a clear and detailed prompt to describe what they want, but I prefer a lazier approach. I like to see what the agent will come up with on its own and then iterate from there.

Here’s the prompt I used.

I want to build a static website for my personal brand, Casandra Campbell. It should have the tagline: Big ideas, sharp analysis. Real-world results.

And here’s what V0 came back with.

Not a bad start! I’m guessing it got the context for the dummy copy from the tagline in my prompt, or maybe it Googled me. I’m not sure!

From here, I got to work editing. I love to work in an iterative back-and-forth fashion, but this isn’t feasible when you’re working with a developer.

For example, I can ask V0 to move something around, look at it, decide I liked it better the old way, and ask V0 to move it back—all in a matter of minutes. If I tried to work this way with a developer, not only would it take ages, but they would probably fire me as a client!

This is why I love vibe coding. It makes it easy to just go with the flow and vibe your way to a polished product, no plan needed.

So this is how V0 and I worked over the next couple of hours, adding content and tweaking things as we went, finally iterating into the website of my dreams. It’s clean, crisp, and professional, and it tells my story the way I want it to be told.

It’s truly amazing that I could do this in a few hours without touching a single line of code or hiring any professional design or development support!

Once it was ready, all I had to do was point my domain name to V0 and deploy it.

It’s live, and you can check it out for yourself: https://www.casandracampbell.com/

You Can Do This Too

It used to be the case that rebuilding my website would take a weekend (or three), or a developer and a lot of decision fatigue. Instead, it took a couple of focused hours, no code, and zero stress.

Vibe coding with V0 gave me a creative playground where I could move fast, follow my instincts, and iterate until it felt just right. No overthinking, no perfection paralysis, just a smooth, modern site that finally reflects where I’m at now.

If you’ve been putting off your own personal site (or any small project) because it felt like too much work, it might be time to try vibe coding. The tools are here, and they’re finally good enough to help you build something great, fast.

Want to see more vibe-coded products? You’ll love this list! 👇

To endless possibilities,

Casandra

Can you do me a favor and tap the ❤️ below? It’ll help me out a lot!

Subscribe for weekly insights that help you start smarter and scale faster.

Subscribe now

You don’t need to know how to code to build software anymore.

Thanks to tools like v0, Replit, and Cursor, anyone with a good idea and a half-decent prompt can spin up a landing page, prototype a tool, or even launch a micro-SaaS with zero technical background.

It’s fast, it’s fun, and it’s totally changing how digital products get made.

But here’s the thing: just because you can build something doesn’t mean you’ll build it right. And if you’re not careful, it’s easy to end up with a broken app, exposed data, or a surprise cloud bill that makes you rethink the whole thing.

So, I asked , a seasoned software engineer and the author of , to break it down for us.

Below, he shares the vibe coding best practices you actually need to know—especially if you’re a marketer, creator, or solopreneur using AI tools to build your first product.

Vibe Coding Best Practices: How to Build Fast Without Breaking Everything

If you’ve spent any amount of time online in the last few years, you’re probably already familiar with how quickly AI tools are changing work. One of the biggest changes is what it looks like to make software. AI-augmented software development unlocks tons of opportunities—and not just for software engineers!

Sure, using AI tools to write code changes the game for software engineers like me, but it doesn’t stop there.

Tons of people with no coding experience at all are using AI tools to build software that makes them money. Many online are calling this vibe coding because the focus has quickly shifted from writing code to building products.

Sometimes, it’s just people scratching their own itch, making software that helps them. Other times, it’s people making full-on SaaS products that they sell to their audience.

Take Jack Fricks, for example. With no coding experience, he built a handful of applications, including one that received an acquisition offer for one million dollars!

We are at a unique window in time where it’s never been easier to write software for your business. Still, vibe coding might be a little daunting if you’re not a software engineer.

But that shouldn’t stop you!

Today, I’m going to walk you through how to get started with some of the best AI tools while staying safe.

Getting Started With Vibe Coding

AI coding tools fall into one of two major categories. The first is AI-assisted code editors, like Cursor or Windsurf. These tools are popular among software engineers, but not as useful for complete newcomers.

If you’re just looking to get a visual application onto the internet, you’re probably going to have the most luck with Loveable, Bolt, or v0.

Among these, my favorite choice to quickly prototype software products is v0.

This is where things really get fun. For our example, I’ve been itching to make a landing page for a digital product I’m working on. Here’s the prompt I started with:

Make a super simple, minimal, sleek sales page for a digital product called "The software engineers quick start guide to using AI for real work". Write basic copy for the product, including a table of contents. Give them a button to buy that is very clear..

V0 will do some reasoning and then kick off the project.

Within 5 minutes, v0 had a prototype that I was already pretty happy with.

Now, what about getting our project on the internet?

Getting Your AI-Coded Project Online

The great thing about using v0 is that all you have to do to get your project on the web is click “Deploy”. In less than ten minutes since I started, the sales page is already on the internet!

The “Buy Now” button doesn’t go anywhere yet, but we’ll get to that later when we talk about payments and Stripe.

So what’s happening under the hood here? V0’s deploy button sets your project up on Vercel, a popular platform for hosting web applications. You can, of course, set up your own server, but using a platform like Vercel abstracts away much of the hassle and risk of running your own infrastructure.

If you’re following along, you’ll probably notice the web address (aka “domain”) of your project is a little weird. Fear not! Vercel allows you to use a custom domain from any domain provider for your project after you get it running.

How to Stay Safe While Vibe Coding

Just because you can write code with AI doesn’t mean everything you produce will be safe. Software engineering is a complicated profession full of pitfalls and gotchas, so you should exercise some caution when working on projects.

Next, I’ll walk you through some pitfalls to avoid while you’re working with AI to make web projects. While this can’t be an exhaustive list, keeping these things in mind will seriously reduce your surface area for risk.

Static Sites Are Safest

The first thing to keep in mind is that static sites are generally the safest option.

Static sites are sites whose content doesn’t change. The sales page we built above is a great example of a static page. It doesn’t handle user input, user profiles, storing data, or payments, so there’s not too much that can go wrong! If you don’t want to handle some of these things, be sure to read the next section for things to keep in mind.

Make Sure Your Secrets Are Secure

Another thing to keep in mind is to always secure your secrets. As your applications get more complicated, you’ll eventually come across things like API keys that should stay “secret”.

There are two major ways to keep your secrets safe.

• The first is to never put secrets in code that shows on the user's browser.

• The second is to use environment variables to store secrets rather than keeping them in your codebase.

Following these two rules significantly reduces the risk that these secrets fall into the hands of a bad actor. If you are wondering if you secured your secrets, ask your coding tool if there’s anything in the codebase that should be secret!

Set Budgets for Your Platform Vendor

One thing to keep in mind for costs is to set budgets for your platform vendor. Whether you’re using Vercel (the default for v0) or another platform, setting a budget will help you avoid unexpected costs. The free tiers on these platforms are quite generous, so I have a $1/mo budget and never run up against it.

Set Up Rate Limiting or Throttling on Your Application

A more advanced technique to protect against bad actors is to set up rate limiting or throttling on your application. This prevents a single actor from using your application fast enough to cause a “Denial of Service” problem,1 which protects you from outages and unexpected costs. Rate limiting is a pretty complicated topic to cover in detail, but you can get a good start with your platform’s tools built specifically for this!

Sanitize Input from Users

If you’re making an interactive application that stores information from the user, be sure to sanitize any input from users. This protects you from “SQL injections,” a common hack that can corrupt or expose your stored data.

Most modern frameworks (like Next.js, which is what v0 uses) make this pretty easy. If you’re storing user input, just look up “SQL sanitization” in your framework’s documentation. If you’re having a hard time, don’t be afraid to ask your agentic coding tool for help and even pass along the documentation.

Note: If techniques like input sanitization, rate limiting, and environment variables are scaring you off and you’re not comfortable with the solutions your AI tool is providing, then stay in the realm of static sites! Being able to make static sites with near-infinite customization at no cost is a huge unlock for internet solopreneurs.

Use Vendors for Risky Things

Because AI lets you write so much software so quickly, you may be tempted to write logic that a third-party vendor might be better suited for. Here are a few examples.

Stripe Payments for Payments

You could try to handle payments on your own, but it's much safer and easier to use a vendor like Stripe to handle payments for you. Stripe will process the payment for you, and your application will never have access to the users’ payment details, making it impossible to have a data breach.

Stripe has made payments super easy with payment links, their all-in-one solution. Payment links are mostly no-code, but you can lean on your AI tool to help you insert the actual link into an appropriate button or link on your website.

Supabase for Database Hosting

If you really want to persist some data from users, you will need a database. This can get pretty complicated, so consider using a managed database service like Supabase to host your database. This abstracts away many database pitfalls and keeps your data safer. Supabase will provide you with links and clear instructions on how to send data to and from the database using API calls, which your coding tool can help you set up.

Still, working with a database means you have the responsibility of keeping your data safe. You’ll need to protect against a whole host of new potential security vulnerabilities, like horizontal escalation and SQL injection. You effectively mitigate these, particularly by using good resources like the OWASP Top Ten with your AI tools, but be sure to be extra vigilant.

Clerk for User Authentication

If you want users to have accounts, consider using an authentication provider like Clerk to store the authentication information, which takes away another risk vector. Clerk has good documentation for getting started, and this route can be done without storing much data.

Go Out and Build Something!

Vibe coding is more than just a fun trend. It’s a name for a new era. You no longer need permission or an experienced team to put together a website. Of course, software engineering isn’t going away. The more dynamic and useful you want your app to be, the more you’ll need some technical experience.

If you want to be in the loop on how you can use these AI tools to make apps and websites, check out The Augmented Engineer. There’s plenty of useful content here, and this article is a great place to start.

The Augmented Engineer

Start here: Welcome to The Augmented Engineer

Here’s the deal. AI is changing what it means to be a software engineer. I say this as someone who writes software for a living…

Read more

a year ago · 1 like · Jeff Morhous 🦾

A huge thank you to for sharing all this insight!

Be sure to check out for more for more great tips and information, especially if you’re interested in learning more about AI-supported software development.

I hope this post made you feel a little more confident (and maybe even a little excited) about building with AI. You don’t have to be a developer to create something useful, beautiful, and profitable. But a little bit of technical wisdom can go a long way toward avoiding dangerous and costly mistakes.

And when you do end up building something, please share in the comments! I’d love to see what you’re working on.

To endless possibilities,

Casandra

Can you do me a favor? If you found this useful, tap the ❤️ below. It’ll help me and Jeff out a lot!

And for good reason. AI coding has gotten so good in the last few months that the barriers to entry for creating new software products have seemingly evaporated overnight.

Now, anyone can create simple software products like web tools, mobile apps, digital games, Chrome extensions, and landing pages without writing a single line of code.

You simply tell AI what you want, and it gets to work. As the AI shares its work with you, you can ask for changes using natural language and let it figure out how to implement them.

If you’ve never tried vibe coding before, it’s hard to believe it’s real. You may wonder if people are actually creating products this way, and if so, what they actually look like.

That’s where I come in. I scoured the web and found 21 great vibed-coded products, which means that AI built these products based on natural language prompts. They span a range of niches and platforms, showing you what’s possible and maybe inspiring you to give it a shot with your own idea.

Vibe Coded Web Tools 💻

A web tool is essentially a software application or program that you can access and use through a web browser over the internet. These are one of the easiest products to vibe code because you simply publish the website when it’s ready.

1. Find Your Business Bottleneck

Find Your Business Bottleneck is a lead generation quiz that helps entrepreneurs identify their biggest business bottleneck. It’s fully integrated with Kit, so quiz results trigger smart follow-ups and personalized product recommendations—no additional tooling required. Previously, I would have needed to pay for a dedicated lead generation software tool, but I vibe-coded this instead (practically for free). The result? A totally custom funnel that saves me money and fits my stack perfectly.

2. Spryngtime

Spryngtime2 is an e-signature tool (like DocuSign). Unlike DocuSign, it’s totally free to use. It’s also fully compliant with the ESIGN Act and UETA, so signatures collected using Sprygtime are legally binding.

3. PropAnalyzer

PropAnalyzer3 helps real estate investors analyze properties. Instead of spending hours on manual analysis with spreadsheets, they can get the information they need to assess deals in minutes.

4. VerseMind

VerseMind4 is a Bible memorization app that uses first-letter mnemonics, spaced repetition, and progress tracking to help users memorize scripture. A developer who wanted to improve on The Bible Memory App (the leading alternative in this space) created it.

5. LCalculator

LCalculator5 provides calculators with a nice clean interface for dozens of different use cases, including:

• A standard deviation calculator that calculates the standard deviation, variance, and other statistical tests for a dataset.

• A snow day calculator that predicts the likelihood of a snow day based on weather and school factors.

• A swim pace calculator that calculates swimming speed and estimated calories burned based on distance and time.

6. Gud Prompt

Gud Prompt6 aggregates thousands of AI prompts that users can save or sort into collections. They also provide a prompt generator, which helps users expand simple prompts, and a Chrome extension that can easily access saved prompts.

7. Home Ownership Cost Estimator

created the Home Ownership Cost Estimator7 for subscribers of. It’s a full-featured mortgage and homeownership cost calculator that breaks down everything—from monthly payments to lifetime costs, including taxes, insurance, and maintenance. You can adjust loan terms, compare scenarios side-by-side, and even export a PDF summary.

8. GrowthFuel Hypothesis Engine

The GrowthFuel Hypothesis Engine8 helps you sharpen your growth experiments by turning vague ideas into clear, testable hypotheses. You input your current hypothesis (however rough), and it guides you to improve it using structured prompts and proven growth frameworks. It’s like having a growth strategist in your browser, nudging you toward smarter, more focused thinking.

9. Kard Kareem

Kard Kareem9 provides a platform for designing and downloading Eid greeting cards to send to friends and family. They provide starting templates with suggested text that can be easily customized and personalized.

Vibe Coded Mobile Apps 📱

No-code mobile app builders have been around for a few years, but they still require a decent amount of time and effort to create and deploy an app, and users are typically limited to preset templates. With vibe coding, you only need to use conversation to build the entire thing from scratch and deploy it to the app stores.

10. NeuroParent App

NeuroParent10 is a mobile app that provides in-the-moment tools for parents of children with autism and ADHD. It’s designed to help them quickly deal with emotional outbursts and sensory overload with clear, easy-to-follow techniques that parents can implement immediately.

11. PeptideCalc.io

PeptideCalc.io11 is a peptide calculator for peptide therapy users. Peptide therapy involves introducing specific peptides into the body (through injections, oral supplements, or topical applications) for specific health benefits. Calculating the dosage requirements for a given protocol can be somewhat complicated, but this calculator makes it easier.

12. Scoreboard

Scoreboard12 is a mobile for keeping score across a range of activities. It can be used for board game nights, rounds of golf, and even keeping track of sets and reps at the gym.

Vibe Coded Chrome Extensions

Chrome extensions are small software programs you can add to your Chrome browser to gain extra features and customize your browsing experience. Some people have started using AI coding to build Chrome Extensions, and we’ll likely see more soon.

13. eli

eli13 is a Chrome Extension for people who love Reddit. In just one click, it summarizes exactly what a Reddit thread is about, so you don’t have to read the whole thing or ask to figure it out.

14. Currency Converter

Currency Converter14 makes it easy to understand the price of online purchases in your own currency before buying. When it’s enabled, it automatically converts prices on webpages to your selected currency.

Vibe Coded Games 🕹️

Digital games that run on websites and mobile devices used to require a lot of design and development skills to create. But with vibe coding, creating your own game from scratch is more accessible than ever.

15. Tetdle

Tetdle15 is a game that combines the mechanics of Tetris with word games like the New York Times’ Spelling Bee.

16. Scene Pixels

Scene Pixels16 is a movie guessing game where players must correctly guess the movie based on a pixelated version of a movie scene.

Vibe Coded Landing Pages 📃

Even simple landing pages can be created through vibe coding. In the past, these would have been custom-coded manually or created using limited templates. But now, fast and creative designs are accessible to everyone.

17. AI for Non-Techies

AI for Non-Technies17 is a newsletter that helps demystify AI using simple language anyone can understand. This newsletter signup landing page was created by vibe coding.

18. PhotoGuru

PhotoGuru18 is a tool for generating a variety of AI photos, like headshots, based on submitted selfies. Their landing page, which looks to be well-optimized for conversion, was created with vibe coding.

19. Build That Idea

Build That Idea19 is a platform for building and monetizing AI agents. Their landing page, which is pretty much the Platonic ideal of a high-converting landing page, was vibe coded.

20. Guillermo Flor Personal Website

built a personal portfolio-style website20 entirely using Lovable, an AI website builder. It features clean sections for writing, projects, and tools, all with a modern, friendly vibe that feels instantly approachable. What’s impressive is how polished and well-organized it is, despite being built from natural language prompts. It’s a great example of how creators are using AI to build beautiful, functional personal sites without touching a line of code.

21. My Personal Website: CasandraCampbell.com

I recently rebuilt my own website, casandracampbell.com, by vibe coding it with V0. I was delighted by how easy it was and couldn’t be happier with my end product!

Are you looking to try your hand at vibe coding? Check out my guide for suggestions, recommendations, and potential issues to avoid.

If you vibe code a product yourself or know about an interesting vibe coded product not listed here, I’d love to hear about it!

To endless possibilities,

Casandra

Subscribe now

When I first got into startups 15 years ago, building new software products was hard.

If you had a startup idea, you either needed to know how to code or you had to find a technical cofounder: a developer with the technical skills to build your product idea and who was willing to do it for equity. In other words, they would work for free until your startup was successful.

Since most startups fail, and most developers know that, it was pretty hard to find a technical cofounder.

Even when you could find a technical cofounder, it took time to build a product that was good enough to launch so that you could validate your product.

And while recent advances in no-code tools have made it possible to launch without a technical founder, they still require a lot of work and you’re mostly limited to templates they support.

Vibe Coding Just Changed Everything.

Now, for the first time, anyone can create a unique software product in a matter of hours, sometimes even minutes.

We're in a golden window of opportunity, when early adopters of vibe coding can stake out a lot of digital land before everyone else catches on. A brief moment where vibe coding is a superpower few possess.

Vibe coding is the great equalizer, allowing small teams and individuals to achieve the scale of large enterprises. While others are still debating, you can build the next generation of AI-powered software products.

Become a digital alchemist, turning ideas into reality with the power of AI coding.

What Is Vibe Coding?

Vibe coding involves describing the desired functionality of a program in natural language rather than writing traditional code. The term "vibe coding" highlights a more intuitive and less rigid approach to development. It signifies a willingness to let AI handle the technical details while you focus on higher-level concepts and ideas.

In other words, it’s perfect for marketers, solopreneurs, small teams, and anyone who needs to prioritize speed over high precision.

The term “vibe coding” blew up overnight after this tweet went viral.

However, vibe coding only became possible in the last few months after rapid improvement in the ability of LLMs to turn natural language into functional code that matches what the prompter is looking for.

What Are the Main Benefits of Using Vibe Coding?

Vibe coding has the potential to democratize software development, making it accessible to individuals with less formal coding training. It can also accelerate development cycles and increase productivity for people who have coding skills.

• With vibe coding, you can create web products faster and cheaper.

• There is less need for extensive manual coding and debugging.

• Anyone can vibe code, even if you don’t have any coding skills.

You know that idea you’ve been sitting on forever? Now, you can work with AI to build it in minutes to hours, basically for free.

Or maybe you have a bunch of ideas you’ve been sitting on. The cost of vibe coding new software products is so low that you can build prototypes for all of them and see what sticks.

What Are the Limitations of Vibe Coding?

AI hasn’t completely replaced developers yet. There are real limits to vibe coding that you need to be aware of.

• Security and Privacy Risks: AI-generated code can introduce security vulnerabilities if not thoroughly reviewed. Don’t create products in high-risk or sensitive industries using vibe coding (e.g., don’t create a platform for therapists to meet with their patients virtually and store their patient notes).

• Complexity and Scalability: AI-generated code may lack the architectural foresight that experienced human programmers provide, leading to inefficiencies and difficulties in scaling. If your product gets big or complex enough, you’ll eventually need to hire an experienced programmer to take over.

• Debugging and Error Handling: While AI can generate code, it often struggles with complex debugging. Identifying and fixing intricate errors, especially in large-scale applications, often requires human expertise, so you may have to make some compromises with how your product is built.

Although AI-generated code has some limits, you don’t need to let these hold you back from trying vibe coding. In the post below, a software engineer shares vibe coding best practices for reducing risk, especially as a non-technical person.

You don’t need the first version of your product to be perfect. It just needs to be good enough to launch. If you get traction and your idea proves successful, you can start working with a human programmer to improve your product as needed.

Plus, AI-generated code has improved dramatically in the last six months, and we have every reason to believe it will continue to improve rapidly.

Just look at the progression of images generated using 2D diffusion models shared by . The improve from V1 to V5 is shocking. It will be the same with AI coding.

If you start now, you’ll learn how to prompt AI to build what you want, and you’ll already be an expert by the time AI coding tools get really good.

Vibe Coding Tools

There are several AI coding tools on the market already that you can use to vibe code:

• Replit: The best option for most.

• V0: Best option for simple products with beautiful UIs.

• Cursor: Best for developers or people who want to learn coding.

• Tempo: Best for developers who want to build React apps faster.

• Windsurf: Best for technical founders who want to write real code—without the boring parts.

• Bolt: Best for designers who want to build from a Figma file.

• Lovable: Easiest for simple projects and tech newbies.

I currently recommend Replit for most people.

Replit offers the best of all worlds if you're not a developer. You can create and deploy products entirely from natural language processing (aka prompting) without touching any code. But you can still access the code if you want to. This means you can bring in human engineers to solve specific problems or deal with increased complexity in the future if you need to.

Aspiring developers should use Cursor.

Cursor was created as a copilot for traditional coding. It can provide a lot of assistance, but you’re still the lead developer on the project. While Cursor can write code for you, you’ll need to learn how to host your products so you can launch them to the public.

What Are Some Real-World Applications of Vibe Coding?

Vibe coding works really well for several types of projects:

• Mobile apps that can be deployed to the Apple App Store and Google Play Store.

• Custom landing pages, including waitlist pages.

• Simple web tools for generating traffic and leads, like this ad spend calculator built using Replit and Claude.1

• Simple web products like directories and job boards.

• Affiliate marketing websites that monetize content with links to relevant products.

• Early versions of complex products that need to be validated with customers before hiring a team to build the full product.

• Basic websites that you can rent or sell to local businesses, like the one created in this video.2

Each of these can be built incredibly quickly using vibe coding, and as the quality of vibe coding improves, more and more will be possible.

Good Vibes Only

Vibe coding is a revolution that is democratizing software creation. While limitations exist, rapid AI improvements mean now is the time to experiment. Early adopters will gain a huge advantage by becoming experts in prompt-driven development and leveraging the first-mover advantage on new products.

Ready to become a digital alchemist and turn ideas into reality with vibe coding? All you need to do is pick an idea, open up Replit or another AI coding tool, and start prompting!

To endless possibilities,

Casandra

PS. If you found this useful, please tap the ❤️ below. It helps me out a lot!

There are so many places online where we’re expected to have photos of ourselves, such as our LinkedIn profiles or our bios on speaker pages for conferences.

For those of us building businesses attached to our personal brand, the expectation is even higher. People typically want to see us in action, and the more we integrate ourselves into our content, the better it performs.

But getting professional photos taken is a big undertaking that often looks like this:

• Find a good photographer and pay a lot of money to book them.

• Stress about choosing the right outfit and spend money on a new one.

• Spend time and money on personal grooming like a haircut or makeup.

• Spend several hours getting photos taken, and if you’re like me, feel really awkward the entire time.

• Wait several weeks for the photographer to edit the photos and send them to you.

What if, instead, you could get professional photos in a couple of hours for only $35?

This is the promise of Aragon.ai! I decided to try it out. Keep reading to see my results!

Table of Contents

• How Aragon.ai Works 🤖

• My Aragon.ai Headshot Results 🎞️

• Is Aragon.ai Worth It? ⚖️

How Aragon.ai Works 🤖

Aragon.ai is a tool that generates brand-new, professional-looking headshots based on the “real” photos you submit.

It’s quite highly rated, and I’ve frequently seen it recommended as the most realistic AI headshot tool.

According to their homepage, they’ve already generated 25 million headshots, which likely represents around 250,000 customers.

And their Trustpilot rating is Excellent with an average of 4.7 out of 5 after 285 reviews.

Given how much I loathe getting my photo taken, I knew I had to try it!

To generate photos, you first fill in a survey that asks about your gender, age, eye color, hair color, and ethnicity. This probably sets some parameters to ensure the generated images don’t get too far off track from what you look like. It may also add some parameters you don’t want, as I’ll point out later.

After you fill in the survey, you’ll need to upload 6-10 photos of yourself. Make sure you read the photo requirements for the best results. No hats, tank tops, silly faces, weird angles, etc.

I uploaded a mix of professional and everyday photos. The only rule I ignored was no tank tops. I almost always wear a tank top and a hat, so it was hard to find enough photos that met all the requirements. I figured the tank top shouldn’t affect how my face is generated, but a hat or side view would. I also hoped it would influence the outfit choices for headshots. Spoiler: It didn’t.

Here are a couple of the photos I uploaded so you can see what I actually look like. Professional photographers in Toronto took both. (I highly recommend them both, so let me know in the comments if you need a referral.)

After you upload your photos, you’ll be prompted to choose a package. I went with the $45 package to get more photos and variety. When I got my results, I was given 100 total photos (the 60 I paid for plus 40 bonus photos).

It probably doesn’t cost Aragon.ai much to generate more photos than you paid for, but it reduces the likelihood you’ll be unhappy and want a refund, so it’s a great tactic on their part.

The final step after paying is to choose your outfits. There are three options: professional, casual, and business casual. I chose all three to see all the options.

Your photos are supposed to be generated in 1-2 hours, but mine were ready much faster.

My Aragon.ai Headshot Results 🎞️

I was initially pretty disappointed with my results, but in retrospect, I probably set my expectations too high after hearing others talk about Aragon.ai. I did get a few photos that I could see myself using, which I guess is pretty good value for $45.

For example, this one looks like a very touched-up photo of me. The teeth are slightly off if you zoom in, but most people wouldn’t. My hair also looks pretty natural, and I could see myself wearing this shirt.

However, most of the images don’t look like me at best…and look extremely fake, at worst.

This one looks like a completely different person.

And this one doesn’t look like a real person at all!

Even the ones that look like me have issues:

• It removed my freckles. Sometimes, they’re not as obvious in photos depending on the light, but it’s kind of crazy for such an identifying detail of my appearance to completely disappear. I guess this is too hard for AI to generate accurately right now.

• It added a pushup bra. As someone with an athletic build, it’s pretty crazy that many of the photos appear as though I moved up a cup size and am wearing a pushup bra. Not only is this completely different from the photos I submitted, but it’s also not what’s in style right now.

• It struggles a lot with teeth. My teeth look really off in many of the photos, and from reading reviews, this seems to be a common experience. The AI clearly struggles to generate this.

• It only generated open-mouth smiles. I take a lot of photos with both open and closed-mouth smiles. I made sure to submit both so that I would get both back, but every single image had me smiling with my mouth open. This is kind of surprising, considering how much the tool struggles with teeth!

• The outfits are ugly. I really don’t like the clothing in most of the photos. Even if otherwise accurate, it’s hard to like and want to use an image of yourself when you don’t like the outfit. There should be more input into the clothing choices since this is probably pretty easy for Aragon.ai to generate.

• They look like glamour shots. All the images look extremely touched up, even if they’re otherwise supposed to look casual. They remind me of the glamour shots teens used to get at the mall in the 90s.

Here are a couple of examples of almost useable photos.

This one looks mostly like me, but I would never wear this shirt.

This one almost looks like me, and while I wish I had these cheekbones, I wouldn’t wear this bra. And if I’m being honest, I probably wouldn’t wear this shirt either.

Is Aragon.ai Worth It? ⚖️

If you don’t have any professional photos or you just want to play around, then maybe.

But another option you should consider is editing selfies to look more professional using Nano Banana. It will look more realistic, and you’ll have more control over the results.

This was a fun experiment, and I will definitely repeat it in the future to see how the technology improves.

I’m personally unlikely to use any of the images I generated, but I’m also lucky enough to have a bunch of professionally shot photographs I can pull from when needed.

If you don’t have professional photos, this is a very easy and cheap way to get some. Out of a hundred images generated, you might only get a handful that are good enough to use, but that’s really all most people need.

If you do decide to try it yourself, I’d love to hear about the results! Come back and share your experience in the comments so others can learn from it, too!

To endless possibilities,

Casandra

PS. If you found this interesting, please tap the ❤️ below. It helps me out a lot!