I was looking at my credit card statement this morning and I realized I am essentially a walking museum of software I do not use. I have subscriptions for things I signed up for in twenty twenty-three that are still just pulling twenty dollars a month from my soul. It is a subscription graveyard, Herman. I have got project management tools for projects that ended two years ago, a specialized image upscaler I used exactly once, and three different note-taking apps that are all currently empty.
It is the SaaS tax, Corn. We have been conditioned over the last decade to believe that if we have a specific problem, the only solution is to find a vendor, sign up for a tiered plan, and give them a permanent seat at our table. We have been renting our productivity instead of owning it. I am Herman Poppleberry, by the way, for anyone joining us for the first time, and I have spent the last seventy-two hours looking at why that entire model is currently on fire. We are seeing a massive shift in the power dynamic between the person with the problem and the person with the code.
Today's prompt from Daniel is about exactly that. He has been looking at this shift from buying those bloated SaaS subscriptions to just building your own bespoke tools with AI. He actually built a custom tool himself recently for translating and formatting Hebrew emails using Claude, and it only took him a few hours. It really gets to the heart of this idea that software is becoming a capability you manifest rather than a product you procure. It is the end of the generic era.
That Hebrew email case study is a perfect example because it addresses a niche, high-friction problem that a generic SaaS company would never prioritize. If you are working with Hebrew, you are dealing with right-to-left text direction, or RTL, which is a nightmare in standard western-designed email clients. You get these weird formatting bugs where the punctuation ends up on the wrong side or the layout breaks when you mix in English words. It is a specific kind of digital friction that makes your professional life just a little bit worse every single day.
I have seen that. It looks like the text is trying to escape the box. It is a mess. And the thing is, there is no "Hebrew Email Fixer" app on the App Store that actually works, because the market for it is too small for a traditional startup to chase.
And if you are Daniel, living in Jerusalem and dealing with this daily, you have two choices. You can wait for a major provider like Microsoft or Google to perfectly optimize their RTL support, which could take years, or you can build a wrapper. Now, in the old world, building a wrapper meant setting up a development environment, choosing a framework, handling the API calls to a translation service, and then manually debugging the CSS for the right-to-left rendering. That is weeks of work for a developer. You would need to understand Unicode-Bidi properties, the difference between logical and visual ordering, and how to inject specific CSS injections into an email header.
But Daniel did it in an afternoon. How does that happen without a team of engineers?
He did it using the new interactive apps feature Anthropic just pushed to Claude on March twenty-fifth. He basically described the problem to the model, gave it some examples of broken formatting, and the AI generated a functional, shareable tool. It handles the translation logic, it applies the correct Unicode-Bidi properties in the CSS to ensure the text flows correctly from right to left, and it gives him a clean interface to work in. This is what people are starting to call vibe coding. You are not writing the syntax; you are describing the intent and the AI is acting as the glue between the API endpoints.
Vibe coding. That sounds like something I would do while taking a nap. But it is actually a pretty radical shift in how we think about labor, right? You are moving from the engineering of the solution to the description of the requirement. It reminds me of what we talked about in Episode eight hundred and sixty-four, where we first saw the cracks in the SaaS model. Back then, we were just speculating, but now it is happening in real-time.
It is the democratization of the software stack. You no longer need a computer science degree to build a functional tool. That is the core of the Agentic Workflow pattern we are seeing dominate this year. When you look at what happened on March eleventh with the release of Replit Agent four, it becomes even clearer. We are no longer just talking about a chatbot that writes a snippet of code. We are talking about parallel AI agents that handle the frontend, the backend, and the database architecture all at once. Replit is now valued at nine billion dollars because they have forty million users who are increasingly non-technical people building full-stack applications.
I saw a report from Gartner that said by the end of this year, citizen developers will outnumber professional developers four to one at large enterprises. That is a massive demographic shift. It is like everyone suddenly got a personal engineering team that works for the price of a few API tokens. And those tokens are getting cheaper. As of this month, March twenty twenty-six, the cost of token-based API execution for standard business logic has dropped by about eighty-five percent compared to where we were two years ago.
That eighty-five percent drop is the tipping point, Corn. This makes the build versus buy math look completely different. If I am a mid-sized company and I am paying a hundred thousand dollars a year for a CRM like Salesforce or HubSpot, but I only use five percent of the features, why would I keep paying that? In the past, the answer was because building a custom CRM was a million-dollar project. Now, you can point an agentic workflow at your existing database and say, "Build me a dashboard that only shows these three metrics and lets me send these two types of emails."
But what about the maintenance? That is always the big scary monster in the room. If I build my own CRM, I have to fix it when it breaks. If I buy Salesforce, it is their problem. Or at least, that is the theory.
That is the big misconception we need to bust. People think bespoke software is a liability because of maintenance debt. But with modern AI agents, the code can be self-healing. This is something we touched on in Episode fourteen hundred and forty-seven when we discussed programming agents in plain English. We are seeing these agents write their own unit tests and run their own diagnostic loops. If an API endpoint changes—say, Stripe updates its checkout logic—the agent detects the failure, reads the new documentation for the API, and updates the connection logic automatically. The maintenance is becoming as automated as the creation.
It is like having a car that changes its own oil and rotates its own tires while you are sleeping. But let's go back to Daniel's Hebrew tool for a second. The thing that strikes me is the specificity. Generic software has to be everything to everyone, which means it is often mediocre for someone with a specific need. Daniel's tool is a sharp knife for a specific task. It doesn't have a "pro" version or a "marketing automation" tab he doesn't need. It just fixes the Hebrew.
That is the unbundling of SaaS. We spent ten years bundling everything into these giant platforms, and now we are seeing them get ripped apart into modular, AI-native internal tools. Instead of buying a seat in a generic CRM, companies are starting to buy capabilities through APIs and then using something like Claude Code to stitch them together into a workflow that only tracks the metrics they actually care about. You are moving from buying an interface to buying an outcome.
And if the value is in the outcome, then the interface becomes a commodity. This is why the traditional SaaS vendors are seeing so much stock volatility right now. If I can generate a visual UI using Replit's Infinite Design Canvas in ten minutes, then your polished, branded dashboard is not worth the premium you are charging for it. I can have a dashboard that looks exactly how I want it to look, with the buttons exactly where I want them.
There was a report on CIO dot com yesterday that said mid-market firms are replacing niche SaaS subscriptions with homegrown tools to save seventy to eighty percent on licensing costs. That is not just a small optimization. That is a massive transfer of wealth from the software vendors back to the businesses themselves. It also gives them back their data sovereignty. When you use a third-party SaaS, your data is often locked in their proprietary silo. It is a nightmare to export, it is formatted in a way that makes it hard to use elsewhere, and you are essentially a tenant on their land.
That is a huge point. If you build the tool, you own the database. You own the logic. You are the landlord. You aren't worried about a company going out of business or changing their terms of service or doubling their prices overnight. You have self-reliance. It feels very frontier-like, doesn't it?
It is the ultimate conservative approach to technology, really. It is about ownership. You are not dependent on a Silicon Valley giant to decide which features you get to use. You are building your own digital homestead. But we have to talk about the risks, because it is not all just saving money and fixing Hebrew emails. There is a dark side to this democratization.
I knew there was a catch. It can't all be sunshine and custom dashboards. What are the IT departments saying about this?
They are terrified. They call it Shadow AI. When any employee can prompt a functional application into existence in an afternoon, you end up with thousands of unvetted, unmanaged apps running inside a company. These apps might not have centralized security audits. They might be leaking data to models in ways the company has not approved. Gartner is actually predicting over two thousand legal claims related to death by AI by the end of this year, mostly due to a lack of guardrails in autonomous software.
Death by AI sounds a bit dramatic for a Hebrew translation tool, Herman. Are we talking about the machines rising up, or just bad code in a medical device?
It is the latter. Well, we do not say that word, but you are hitting the nail on the head. The black box decision-making is a real concern. If the AI writes the code and the code has a logic error that leads to a financial loss or a safety failure, who is responsible? Is it the person who prompted it? Is it the company that made the model? Is it the platform that hosted it? The legal framework is still trying to catch up to the speed of the technology. We are in a "move fast and break things" era, but the things being broken are now core business infrastructures.
And then there is the skill atrophy. If we are all just prompting and no one is actually learning how to debug or understand the underlying architecture, what happens when the AI hits a wall? We have talked about the Agentic Throughput Gap before, where the AI just gets stuck in a loop. If you do not have the critical thinking skills to step in and fix it, you are helpless. You are a wizard who has forgotten the language of the spells he is casting.
Gartner warned about this too. They are suggesting that by next year, we might see companies implementing AI-free skills assessments during the hiring process. They want to make sure that if the lights go out, you still know how to build a fire. There is a real fear that we are losing the ability to understand the tools we are manifesting. If you can't explain why the Hebrew text is now rendering correctly, do you really own the solution, or are you just renting a miracle from Anthropic?
It is a weird paradox. We are more powerful than ever because we can build anything, but we are also more fragile because we might not understand how any of it actually works. But for someone like Daniel, the trade-off is clearly worth it. He has a specific, recurring pain point. He has the technical literacy to guide the AI. And now he has a tool that makes his life in Jerusalem easier every single day. He is not waiting for a permission slip from a software vendor. He just solved it.
That feels like the real takeaway for me. We need to stop looking at software as this thing we have to beg for. We need to start looking at it as something we can just command. If you have a workflow that is slow, or a formatting issue that is annoying, or a data set that is messy, you don't need a subscription. You need a good prompt and a couple of hours. The "Prompt-as-Code" evolution means that your description of the problem is becoming the source code itself.
So, how does someone actually start? If I am sitting here with my subscription graveyard, what is my first move?
I think people should start by auditing their own graveyard. Go through your bank statement, look at every SaaS tool you are paying for, and ask yourself: "What is the one feature I actually use here?" For most people, they are paying for a Swiss Army knife when they only ever use the toothpick. Once you identify that one feature, try to describe it to a model like Claude or a tool like Replit Agent. Say, "I want a simple web app that does exactly this one thing."
That is the "Build vs. Buy" matrix for the AI era. If the tool is a commodity—like email hosting or cloud storage—just buy it. But if the tool is your competitive advantage, or if it solves a specific pain point that makes you unique, you should probably build it. You want to own the logic that makes you special.
Start by automating one specific, painful workflow. Don't try to replace all of Salesforce on day one. Just build a tool that handles your specific lead intake or your specific Hebrew email formatting. Once you see how easy it is to maintain and how much money you save, the momentum builds itself.
I am going to do that this afternoon. I suspect I can replace about half of my monthly bills with a few well-crafted paragraphs. I might even have enough money left over to buy you a nicer hay bale, Herman. I know you have been complaining about the quality of the forage lately.
I would appreciate that. But let's look at the broader implications for the future. If we move toward a world where everyone has their own bespoke stack, what does that do to the concept of a personal operating system?
It means the operating system is no longer a collection of apps. It is a single, unified agent that generates the interface you need in the moment. If I need to schedule a meeting, it doesn't open a calendar app. It just builds a scheduling interface, handles the logic, and then the interface disappears when I am done. The OS becomes a shape-shifter. It is just-in-time code.
That is the ultimate vision of SaaP, or Software as a Prompt. The software doesn't exist until you need it. It is a fascinating transition, and honestly, it is one of the most empowering shifts I have seen in my career. We are moving from being consumers of digital products to being architects of digital solutions. A16Z released a whitepaper earlier this month called The End of the Seat-Based License, and they hit on this exact point. They pointed to companies like Intercom, whose Fin AI now charges ninety-nine cents per successful resolution rather than a flat fee per user. The value is moving from the interface to the outcome.
Even if those outcomes are just for fixing Hebrew email formatting. It starts there, but it ends with a complete restructuring of the global economy. No big deal. Just another Tuesday in twenty twenty-six, Herman.
It really is. It is a brave new world of bespoke buttons and custom code. I am looking forward to seeing what people build when they realize they don't have to wait for a feature request to be approved by a product manager in San Francisco.
It is a lot to take in, but I think the empowerment angle is the right way to look at it. We have spent so long being told that technology is this complex, arcane thing that only a few high priests in Silicon Valley can understand. And now, the gates are open. If you can describe your problem, you can build your solution. And that is why Daniel's prompt is so relevant. It is a small, personal story that illustrates a massive, systemic change. It is about taking back control. It is about the democratization of the stack. And it is about the end of the generic era.
I think we have covered the meat of this one. If you are listening to this and you have built something weird or useful or just plain better than the SaaS version, send us an email at show at myweirdprompts dot com. We want to hear about your bespoke tools. We want to know what you are building in your digital homesteads.
We really do. This has been My Weird Prompts. I am Corn.
And I am Herman Poppleberry. Thanks to our producer, Hilbert Flumingtop, for keeping the agents running smoothly behind the scenes.
And a big thanks to Modal for providing the GPU credits that power this show and our research.
If you enjoyed this dive into the death of SaaS and the rise of bespoke tools, do us a favor and leave a review on your favorite podcast app. It really helps other digital homesteaders find the show.
Find us at myweirdprompts dot com for the full archive and all the ways to subscribe.
See you in the next prompt.
Goodbye.
