Skip to content
ChatGPT · Codex · ClaudeEpisode 135 · 17 July 2026 · 31:30

Which Software Will Become Unnecessary? ChatGPT and Claude Are Already Building It

Central question

Which familiar CMS, CRM, forms, bots, and SaaS tools can already be replaced by a system assembled by AI for a specific workflow—and where does custom development remain risky?

What you take away

Understand when it is better to buy an existing service and when to build a custom system around data and business processes—and which limits of models, integrations, regulation, and human review cannot be ignored.

Main threads

What to watch for

1Break every subscription into its function, data, integrations, and maintenance cost; this reveals what the service is actually providing.
2Start with the end-to-end process—not the page or interface: from the incoming user and data to the decision, publication, and analytics.
3Separate reusable components, the data model, and access rules from one-off screens that AI can regenerate.
4Before replacing SaaS, calculate token cost, model runtime, review effort, and the consequences of an error in the real workflow.
Signals to track afterwards
Whether one AI-native system can reliably combine CMS, forms, contacts, analytics, and publishing without constant manual repair.
How pricing and the value of general-purpose SaaS change when basic functions can be assembled for a specific company.
Whether long-running agents preserve context, quality, and predictable cost.
How platforms handle regional requirements for data, cookies, communications, and consent.
Most useful for
founders and digital product ownersproduct, marketing, and operations leadersCTOs and internal-tools developersSaaS foundersteams paying for many disconnected software services

Key takeaways

00:00The Risk Extends Beyond One Product to an Entire Subscription Layer

When a model can assemble a website, CMS, forms, bots, and analytics around one workflow, the familiar stack of separate services is no longer the only way to operate.

04:21Websites Reveal the Shift in the Mass Software Market

A website is not a laboratory demo but a mass-market gateway to the internet. If AI changes how websites are built and managed, the same logic will gradually reach mobile apps, CRM, and ERP.

08:10Long Autonomous Runs Are Both a Capability and a New Uncertainty

GPT-5.6 Sol can work for days and run parallel threads, but duration alone does not prove quality. Scale of work must be separated from slowness and uncontrolled resource use.

11:48Off-the-Shelf SaaS Loses Its Advantage When AI Can Build the Required Function Directly

The choice among CMS, open source, and cloud software changes when Codex and Claude can build a function around specific data without requiring the business owner to inspect code.

14:36Value Comes from Connecting Channels and Data, Not from a Standalone Form

A questionnaire should continue across the website, Telegram, WhatsApp, SMS, or a call with a manager while preserving one context. A standalone form builder solves only a small part of the problem.

21:40A Real Product Must Handle the Rules of Different Markets

A system serving an international audience must account for GDPR, cookies, languages, channels, and analytics. Those details separate a demo from working infrastructure.

25:26The Business Process Does Not End with a Polished Interface

Payments, multiple cards, booking rules, accounting, and exceptions make the real process harder than the page. AI saves development effort only when it captures this operational logic.

29:11The Core Investment Choice Is Not the Technology but the Solution’s Useful Life

Before buying another subscription or building another system, ask whether that layer will still matter in three to five years or become a standard capability of the AI platform.

What this episode is about

A large website, a custom CMS, forms, a contact database, analytics, and campaign-specific landing pages can already be assembled through Codex, ChatGPT, and Claude. This episode is not a “website in five minutes” demo; it is about a shift in the economics of software itself.

The Experiment Started with a Large Live Website
I took a large live website with hundreds of pages and gave the AI systems a practical task: move it into a new environment, understand its structure, preserve the content, and build a custom management system around it. Codex parsed the website, the models helped assemble the architecture, and Claude Fable and GPT-5.6 Sol were used for different parts of the process. This was not a laboratory example or an empty landing page; it involved real volumes of material and real requirements.

AI Is Beginning to Build an Entire Working Environment
A project like this quickly becomes more than a website. The system can create page templates, separate landing pages for products, countries, states, and advertising campaigns, forms and questionnaires, a contact database, internal analytics, bots, and publication rules. Previously, each layer required a separate SaaS product, another integration, and another subscription. A model can now assemble many of those functions inside one system tied directly to a specific business process.

The Main Question Is Not Whether One CMS Disappears
This does not mean every CRM, CMS, and website builder disappears tomorrow. Current models still make mistakes, consume expensive tokens, operate at different speeds, and require human supervision. But the value of the standard software layer is changing. If a business can quickly build its own form, report, or dashboard, it begins to evaluate a generic service, its limits, and its integration cost very differently.

The Investment Is No Longer Only in Software
The episode is therefore less about website creation than about deciding where to invest time, money, and attention. Should you buy another subscription, build a custom system, wait for the next generation of models, or learn to describe the process more precisely? AI can already perform a large amount of technical work, but a person still has to understand the business, set requirements, verify the result, and decide which functions are actually necessary. That becomes the main constraint as code and interfaces become faster to create.

AI does not eliminate software, but it sharply reduces the value of standard intermediary layers. The winning system is built around the real process, data, and accountability—not around another subscription.

Episode transcript

The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 47 segments: 47 identified, 0 mixed, 0 marked with ✓, and 0 unresolved.

Loading…