Monologue: AI Isn't Replacing Software Companies, Calm Down About Claude Code - Better Offline

Better Offline – Monologue: AI Isn't Replacing Software Companies, Calm Down About Claude Code

Host: Ed Zitron
Date: February 20, 2026
Podcast Network: Cool Zone Media & iHeartPodcasts

Episode Overview

In this sharp monologue, tech industry veteran Ed Zitron takes direct aim at the prevailing media-driven panic over AI-generated code—specifically tools like Claude Code—replacing software companies. Zitron dismantles the myth that large language models (LLMs) can feasibly generate functional, secure, and scalable business software, highlighting the complexities and liabilities overlooked in the hype. He also critiques media outlets for uncritically parroting the narratives of AI companies, and forecasts a harsh reckoning when the "AI bubble" bursts.

Key Discussion Points and Insights

1. The Origin of Panic in the Software Sector

Zitron opens by referencing a “dramatic sell off in software stocks,” rooted in fears that generative AI will make SaaS companies obsolete, as companies could “simply build their own software” instead of paying for platforms like Salesforce or Microsoft.
He calls this a “genuinely stupid assumption,” attributing it mostly to analysts and reporters who either misunderstand or ignore the technical realities involved.
- Quote:
  - “In their mind, one can simply type ‘Build me Salesforce now’ into Claude code and have it barf out an identical, functional clone…” (01:01)

2. The Realities of Maintaining Software

Users aren’t just paying for tools—they’re paying for maintenance, compliance, and ongoing infrastructure management.
Even “minor things like currency changes or time zone shifting can cause major problems,” especially if the system wasn’t built with long-term intention and oversight—something LLMs aren’t designed to provide.
Connecting enterprise systems (billing, customer data) involves legal, technical, and security hurdles (e.g., SOC 2 audits), and improper handling exposes companies to litigation and regulatory risk.
- Quote:
  - “…every little fuck up just became your problem. And you’ve got nobody to scream at because while your company is saving $15 per seat per month, you’ve also fired the people whose job it is to make sure your nasty little software subscriptions actually fucking function.” (02:40)

3. AI-Generated Software Cannot Scale or Sustain

The hype ignores operational complexity: who will maintain or debug these AI-coded tools? Most companies would need to reassign engineers indefinitely to support or extend internally generated software.
Zitron mocks the endless corporate urge to “clone” productivity software (Salesforce, Trello, Asana, Notion) with LLMs, describing the logistical absurdity in maintaining dozens of bespoke or partially functional tools.
He underscores the danger: any mistake or oversight is now on the company—not a SaaS vendor’s support team.

4. Media Critique: Who Pushes This Narrative, and Why?

Zitron argues the myth of LLMs replacing SaaS is pushed by people “either fundamentally disconnected from how the world works or actively incentivized to mislead you.”
He highlights how this narrative is uncritically repeated by business media, which “makes [him] genuinely worried that we don’t have a media industry prepared to dissect fundamentally deceptive narratives.”
- Quote:
  - “Please stop seeing every incremental improvement as proof of whatever marketing slop Wario Amadei or clammy Sam Altman is trying to cram down your throats. You’re being played. You’re being conned, and by extension you’re conning your listeners, your readers, and your viewers.” (04:47)

5. The AI Bubble is Heavily Subsidized—and Unsustainable

Zitron’s research on Claude’s API pricing shows that for a $20/month Claude subscription, a user can easily generate $100+ in API compute costs, with similar losses at higher pricing tiers.
- “It takes money to lose money, but this is the reality of the AI bubble. Everything you are using is being subsidized.” (06:07)
The heavy subsidization is intended to “bake [AI tools] into your life,” even though they’re “not good enough” for real-world reliability.
If real costs were charged, casual users hyping Claude Code would be paying “$200 plus dollars a day”—an unsustainable model.

6. Why AI-Generated Apps 'All Look the Same'

Apps built with AI (“Vibe coded apps”) generally lack nuance, vision, and custom functionality; they look similar because they’re regurgitated from the same training data.
AI tools can cobble together superficial clones of existing apps, but anything more “nuanced or unique” is beyond them—because “that’s not what they do…they copy software that’s already been written.”

7. Long-Term Maintenance Nightmares

Zitron projects forward: If, in the future, 90% of a company’s code was written by AI, there’d be no institutional knowledge about why code was written a certain way or how to fix it.
Tracing problems, making improvements, or even understanding operational choices becomes a nightmare:
- Quote:
  - “At some point you’re just looking at a kind of a rat’s nest of crap—of code written unintentionally, spooged out, meaningless. How do you fix that long term?” (07:30)

8. Inevitable Reckoning for Hype and Hysteria

With the current cost structures, Zitron is confident “people are going to drop this shit like a bad habit” once subsidies run out and the true cost is felt.
He predicts both annoyance and vindication: “I can’t wait. I’m gonna be smug about it. I’m gonna be annoying about it. Not gonna lie.” (08:01)
Zitron ends by promising more “hater season” content and teases upcoming interviews.

Notable Quotes & Memorable Moments

On software company value:
- “When you pay a software company even a dog shit, mediocre one, a monthly fee, you're not just paying them to access the software, but to take away the burden of maintenance that comes with running a software company.” (01:41)
On the seductive illusion of LLMs:
- “Just because it’s possible for a non coder to cobble together a website that looks near identical to a model’s training data in the space of an hour doesn’t mean that we’re replacing every software company.” (04:05)
On the media’s complicity:
- “Once the bubble pops, I believe they will demand an explanation from you. I certainly will.” (05:18)
On AI-generated code at scale:
- “What happens in three years or four years when you go and look back on that? You go, shit, why did we do that? Oh, the person left or the person didn’t leave, they’d just been drinking heavily and they don’t really remember what they were prompting the model with.” (07:14)

Timestamps for Key Segments

00:39 – 02:20: Introducing the AI panic, what’s really involved in SaaS value
02:20 – 04:00: The real cost of DIY software via LLMs, compliance, maintenance burdens
04:00 – 05:30: Media’s irresponsible amplification of the AI hype cycle
05:30 – 07:00: Costs of LLMs and the unsustainability of subsidized AI coding tools
07:00 – 08:15: Long-term nightmares of AI-generated business software
08:15 – End: Zitron’s closing thoughts and preview of next “hater season” content

Summary

Ed Zitron’s monologue delivers a reality check for anyone believing the hype that AI tools like Claude Code will make software companies obsolete. Far from being risk-free or scalable, LLM-generated code brings hidden complexity, legal exposure, and maintenance headaches. The media, he argues, has failed to scrutinize the AI gold rush and is setting up both itself and the public for disappointment. Once the subsidies dry up and reality intervenes, many will realize that professional software companies provide much more than just code—they deliver trust, support, compliance, and continuity. Zitron’s message: calm down, stop believing every AI press release, and brace for a reckoning in the over-inflated AI software market.

Transcript

A (0:00)

This is an iHeart podcast. Guaranteed Human.

B (0:05)

At Oppenheimer, we're proven because we're grounded in discipline. For 145 years we've been building and protecting wealth through every market cycle with precision, clarity and the courage to think boldly beyond the moment. This is what market tested legacy looks like for this generation and the next. Put the power of Oppenheimer thinking to work for you. Wealth management, capital markets, investment banking, Media.

C (0:39)

Hello and welcome to this week's Better Offline monologue. I'm your host, Ed Zirtron. Better Offline so in the last few weeks there's been a dramatic sell off in software stocks driven by the anxiety that companies will, instead of paying for someone like Salesforce or Microsoft, simply build their own software. It is a genuinely stupid assumption based on analysts and reporters that simply do not care about the truth. In their mind, one can simply type Build me Salesforce now into Claude code and have it barf out an identical functional clone that's compliant, secure, stable, all because somebody was able to bonk it on the head enough times to spit out something that sort of looked like a tool like Trello maybe, or a personal website. Look, when you pay a software company even a dog shit, mediocre one, a monthly fee, you're not just paying them to access the software, but to take away the burden of maintenance that comes with running a software company. Minor things like currency changes or time zone shifting can cause major problems in systems that aren't built with intention. You know, like something an LLM would spit out. And things get even more complicated when you start connecting other systems of record like billing or a customer's personal information, especially, especially if they're in Europe, by the way, plan to have any of that information actually connect with customers systems. Well, you're going to need a SOC 2 audit and you're definitely going to need to make sure it's got rock solid security so that nobody can swipe all of that data and then you get sued. I also assume you're going to effectively take an engineer off of one of your teams, probably for good, to be honest. To maintain your new Internal Salesforce Monday, Microsoft 365 and Notion clones. Your Trello clone as well, probably your Asana one. I mean, how much software are you going to build? Good thing you've got clawed code to help speed that up, right? Just make sure you read everything it writes because every little fuck up just became your problem. And you've got nobody to scream at because while your company is saving $15 per seat per month. You've also fired the people whose job it is to make sure your nasty little software subscriptions actually fucking function. And while you may fear that a boss might try and force this down your throat to save money, I must be clear how impossible this task is. Even the most annoying, frustrating software as a service contract protects you from the grueling underlying maintenance and infrastructural bullshit that goes into making sure the thing you're paying for actually loads and functions wherever you load it, and even on the browser that you want to load it in in most cases. Thank you, Riverside. The people pushing this narrative are either fundamentally disconnected from how the world works or actively incentivized to mislead you. I've seen this narrative propagated on multiple different business television shows and supposedly outlets, and it makes me genuinely worried that we don't have a media industry prepared to dissect fundamentally deceptive narratives. Just because it's possible for a non coder to cobble together a website that looks near identical to a model's training data in the space of an hour doesn't mean that we're replacing every software company, nor is its ability to do so any indicator that we'll be able to do more than that in the future. I plead with the media, please stop filling in the gaps. Please stop seeing every incremental improvement as proof of whatever marketing slop Wario Amadei or clammy Sam Altman is trying to cram down your throats. You're being played, you're being conned, and by extension you're conning your listeners, your readers, and your viewers. And once the bubble pops, I believe they will demand an explanation from you. I certainly will. And really, in this era, I think people are underselling how big a reckoning there will be when the bubble collapses. How are we meant to trust anybody who vociferously pushed this or who got obsessed with AI once all of this falls apart. When I say falls apart, I mean that the current rates that you are paying are not even close to being sustainable. I'm currently working on a piece called the Haters Guide to Anthropic on my premium newsletter and I found a mathematical study that found that on a $20 a month Claude subscription you can spend over $100 of Anthropics API tokens and on the $200 a month, about 2,300 to $2,700. You got it. I guess it takes money to lose money, but this is the reality of the AI bubble Everything you were using is being subsidized. It's being subsidized by these companies in the hopes that it kind of bakes into your life, except it's not good enough to do that. The frenzied media push under Claude code existed to kind of beguile you, to make you think that there's more happening than a tool that might be able to kind of build a website, sort of, or build you some half functional software that maybe works sometimes. And when it comes to actually building that into something, you could sell your shit out of luck, you actually do need to code. And even the things that you Vibe code aren't really secure or functional. And even that mediocre web slop you're seeing spooged out by Claude code is heavily subsidized. If people had to pay the real rates, those people you see jacking off on Twitter about Claude code, they'd be paying 200 plus dollars a day. Do you think they'd actually pay that? Do you think that's actually happening? Because I'm sick and fucking tired of hearing all the people rambling about Claude code writing all the goddamn code. I'm sick and tired of it because I don't see what the end point is. I don't hear an actual result from this. I don't think people should be. But if this was doing the thing that they were saying it would, that would be happening and it would be definitively connected to AI, or more software would be being shipped that's actually good. Just more software existing doesn't mean the software is functional or useful. And if you look at Vibe coded apps, they all kind of look the same. And that's a result of them all using the same training data. And it's why these things are not really good at building nuanced or unique software. Because that's not what they do. They copy software that's already been written and they do so with no intention, no real plan, and nothing you can look back on and say, oh, right, that's why they built it this way. And perhaps that's not a problem as you're a solo person building your own diddly little app. But expand that to a thousand people or 10,000 people, let's say five, ten years passes and let's take one of these what I think are bullshit statements about how, oh, 90% of our code's written by AI. Great. What happens in three years or four years when you go and look back on that? You go, shit, why did we do that? Oh, the person left or the person didn't leave, they'd just been drinking heavily and they don't really remember what they were prompting the model with. And oh man, how do we actually see why they built this or what the reason was? This becomes a huge problem when things start breaking or when you have to make things more efficient. And then people will say, oh, we use AI to fix that. At some point you're just looking at a kind of a rat's nest of crap of code written unintentionally spooged out, meaningless. How do you fix that long term? How do you build an efficient software package out of that? How do you build software that continues to run well into the future? Is the idea that you just build AI on AI? God, no. Anyway, I think once the real costs are charged, people are going to drop this shit like a bad habit. I can't wait. I'm gonna be smug about it. I'm gonna be annoying about it. Not gonna lie. But I think based on the YouTube comments, you're all gonna join me now. Next week we've got more hater season and I'll be back with one of these monologues maybe. Or perhaps I'll give you multiple interviews. Done some weird ones. Got Victoria Song on wellness, got Adam Becker on Epstein. I mean, got all sorts of shit coming. Looking forward to bringing you more. And I.