rbefored.com - The First Wave of the AI Bubble Bursting My Predictions

Sitemap## Delta CX Hive

“Reflection Before Direction.” Delta CX is about creating & managing change both in work and life. From Customer Experiences in our work to Catalyst Experiences in our own personal improvement, the Hive is all about supporting you wherever you are on your path. Resources: dcx.to

We hear a lot about the AI bubble and how it’s going to burst. I think there will be two waves of bursting.

The second will be companies who realize that the investments aren’t sustainable compared to the outputs and outcomes. They’ll have to adjust spending and investment, and balance AI work with human work. Not ” human in the loop ” as if we’re an afterthought or nice-to-have. Real balance where we still do the work, and AI is a tool that augments us (when correctly selected).

Today, let’s talk about the first wave of AI bubble burst, which I also covered in chapter 11, “Competitors Move Into Your Space,” in my new book, ” Atomic Product-Market Fit."

"We Do X but with AI.”

The first wave of AI bursts that I predict we’ll see will be that all of the products, services, and startups offering X but with AI will burst. If the main PSE (products, services, and experiences) the company offers are layering AI onto X or using AI for some aspect of X, why spend money on that when you can pay for the AI/LLM tool itself, which can do all of these functions?

Examples:

• A startup that takes HR system data and gives you reports and info via AI.
• A startup that helps lawyers review and collaborate on contracts… using AI.
• A startup that uses AI to help you analyze, synthesize, and report on qualitative research.

On the surface, you might think yes, there are real markets for these. These companies will find product-market fit because there will be real demand for what they do.

But We Aren’t considering Two Possibilities

1. If the system that provides the data ever adds these AI capabilities, nobody will need your startup. If the HR system creates better data analysis or reports, your whole startup might crash. Fast loss of PMF.
2. Especially with a paid LLM account (I prefer Claude), I don’t need to pay for that HR startup, legal startup, or research data startup. I can put my data into Claude, and ask it to create exactly what I need done with that data.

Using my own Claude account, I will have more flexibility over what I can do with my data. I’ll have more control over that data. The more companies that see my data, the more I have exposed my data to all of the I-Don’t-Know-Whats in terms and conditions.

This means that “we’re X but with AI” startups get squeezed from both directions. Data sources or integration partners create their own features, satisfying the gap the startup addressed. And users handy with their favorite LLM will move the work into that.

Won’t Somebody Still want what These Startups Offer?

If you threw money at startup X because your HR system didn’t do what you needed with your data… and suddenly, your HR system now offers that… or you figure out how to do it in your job’s approved LLM… you don’t need that startup.

Get Debbie Levitt’s Stories in Your Inbox

Join Medium for free to get updates from this writer.

It’s a lot of friction to export something from one tool and bring it into another. If you’re doing that anyway, why not bring it into your preferred LLM? But if you no longer have to do that because the HR system built the reporting you needed, or built in AI to allow you to create custom reports, then the startup is solving a problem people no longer have.

My Own Examples

• I stopped using Quickbooks to use Zoho Books, but only for last year. This year, I just did my taxes by exporting from all of my bank and credit sources, and having Claude categorize and total everything for me. No special coding, just a chat.
• Last year, I dumped Replit when I realized I could make what I wanted in Claude. And not even Claude Code! I’m not a dev. And not even Claude Opus! I did it in Claude Sonnet 4.x. I built my own private Android app to replace Google Voice so I can run my own VoIP numbers through Twilio. I manually FTPed the code, I ran PuTTY, and I built in Android Studio, all coached by Claude since I only knew FTP beforehand.
• Last year, I dumped Upheal and then Fathom when I realized that I could get a better meeting summary/notes from Claude. I now have Fireflies transcribe the meetings. I give Claude the transcript, and run my usual prompt to get notes in the style I prefer. I could go cheaper and use the built-in transcription for Google Meet calls if I wanted to, but I’m happy with Fireflies’ accuracy and features, which currently offer me more than Google Meet’s AI features offer.

Zoho Books does enough that it’s unlikely to burst with an AI bubble. But for people like me using it for simple bookkeeping that’ll move totals into an annual tax return, the value might not be there.

Replit has Claude as one of the models you can use. Do we still need Replit when we can work directly in Claude? Miro uses Claude, but Miro has other value for me outside of AI prompts.

While working on our AI transformation, we’ll have to look closely at when we want to use the AI/LLM more directly, and when it makes sense to license other tools.

Claude is Becoming a Platform, and it Might Be the only One You Need

Many roads lead to Claude. With chat-based models, artifact creation, the API, Claude Code, and other tools, it’s truly becoming an ecosystem. It’s a smart move. Learn to use Claude decently, and there are so many things you can replace.

Maybe someone else wins or ties Claude. But from what I’m reading, Claude is the closest to having a true ecosystem that’s ready for veteran experts and newbies to use efficiently.

So… Now What?

As chapter 11 of my book reminds you, you will need to closely examine who your target audience is. What they have, what they need, room for improvement in their tasks, and where friction lives now.

There is no product-market fit (or PSE-market fit) without fitting PSE to your market. If your market can take care of these tasks without you, these might not be problems, and therefore nothing to solve.

Invest in fresh and Excellent Research to learn what your target audience needs, and where are the opportunities to deliver quality and value to them. In the near and distant future, it’s unlikely to be, “We do X but with AI.”

Image by Braedon Mcleod on Unsplash.

🐦🔥 Atomic PMF https://atomicpmf.com, Life After Tech https://LifeAfterTech.info, and Delta CX Coaching https://DeltaCX.Coach are part of the Delta CX Hive 🐝, which includes coaching, a free Discord community, articles, and other resources. The Delta CX Hive supports you wherever you are in your life and career paths.

Delta CX has 30 years of experience in helping people brainstorm, start, experiment with, and grow startups and new business ideas. https://deltacx.com

Check out our upcoming live streams and events! https://dcx.to/events including our upcoming cohort course, “AI and Excellent Research for Conversion, Retention, Growth, and PMF,” and it’s free 1-hr Lightning Lesson webinar. https://cxcc.to/courses

🔗 All books, channels, communities, and other resources at https://dcx.to

Last published Mar 23, 2026

“Reflection Before Direction.” Delta CX is about creating & managing change both in work and life. From Customer Experiences in our work to Catalyst Experiences in our own personal improvement, the Hive is all about supporting you wherever you are on your path. Resources: dcx.to

39 following

“The Mary Poppins of CX & UX.” Strategist, Researcher, Architect, Speaker, Trainer. Algorithms suck. Join my community (free) at https://dcx.to

Responses (5)

Write a responseWhat are your thoughts?

The Replit to Claude migration is an important pattern because it crossed a line most people assumed was safe. Replacing a dev tool without being a developer, using Sonnet rather than Opus, is a signal about where the ceiling on direct LLM use actually sits.

3

Totally agree with you. Claude and Claude Code are my "meta-app" right now, meaning that I could simply prompt to create my very own personalized one-time-only application. I used these one-time-only applications for specific tasks and then simply dumped them. The end of domain-specific SaaS is coming...

5

Great article, thanks for sharing. You focus mainly on impacts of AI at the individual level (solo user). Curious as to your thoughts on the sustainability of startups focusing on AI implementation at the enterprise level where management is more complex.

3