TL;DR
Will AI agents replace the CMS? Two perspectives: Thesis — Small companies should work AI-first, websites as code + Markdown, agents handle everything. Antithesis — CMS will remain because adoption is slow and expertise still matters. The truth is probably somewhere in between — and we’ll know in three years.
Contents
- What is the thesis?
- What is the antithesis?
- Where do the differences really lie?
- What does this mean for companies?
- FAQ
What is the thesis?
In short: Small companies that want to work AI-first no longer need a CMS. Coding agents can manage websites directly.
The classic way: Designer designs, agency implements in Webflow/WordPress, editor maintains content in CMS. Every change is cumbersome, requires login, clicks, copy-paste.
The new way: The website is code. Content lives as Markdown files. A coding agent like Claude Code or Cursor can edit everything directly — change text, adjust pages, deploy.
What this looks like:
- I tell the agent: “Change the text on the homepage”
- The agent edits the Markdown file
- I look at the preview
- Good? Agent deploys
- It’s live in two minutes
No login. No CMS interface. No editorial workflow. Just a conversation with the agent.
The cost calculation:
- Classic website relaunch: €30,000–50,000
- With offshore agency: €15,000–20,000
- AI-first setup + consultant: €5,000
The difference: I only need someone to set up the system once. After that, the agent takes over — with support from a working student or interested employee.
What is the antithesis?
In short: CMS systems will remain — especially for small companies that have neither budget nor know-how for the AI-first path.
The reality in German companies: Paper notes, outdated interfaces, processes from 15 years ago. The theoretical possibilities of digitalization have existed for years — they’re barely used.
The arguments:
Adoption is slow. Smaller companies — craft businesses, SMEs — have no IT department. Building infrastructure for AI-first requires understanding that isn’t there.
The Google AdWords parallel: When Google democratized its tools, theoretically everyone could run ads themselves. The result was mediocre. Armies of specialists were hired anyway.
Expertise counts. Designers aren’t artists — they understand information architecture, user experience, business models. An AI-generated website might look okay, but “okay” isn’t enough.
The interfaces remain. Even if AI does the work — there will still be dedicated applications. CMS, CRM, ERP. The way of interaction changes, but the tools don’t disappear.
Where do the differences really lie?
In short: It’s not about technology, but about readiness — and that’s distributed differently.
Both sides agree: AI agents will dominate knowledge workers’ tasks. The question is just: How fast? And who goes first?
The innovators:
- Have someone with IT interest (working student, boss’s son)
- Use ChatGPT, Claude, Cursor as their “IT department”
- Accept that things sometimes go wrong
- Save 80% of classic costs
The majority:
- No IT department, no interest in experimenting
- Trust established systems
- Don’t want to take risks
- Prefer paying more for security
The truth: Both are legitimate. The question is just what fits the company.
What does this mean for companies?
In short: The decision depends on whether someone in the company wants to take the AI-first path.
For AI-first:
- Massive cost savings
- Faster changes
- Independence from agencies
- Learning for the future
For classic CMS:
- Proven processes
- No training needed
- Less risk
- Expertise in-house or with partners
The middle solution: A consultant sets up the AI-first system. An interested employee takes over maintenance. For complex changes, you get help again.
This is the most likely path: Not completely self-service, not completely outsourced — but a new model of collaboration.
FAQ
Does AI-first really work without developers?
For simple websites, yes. You need someone to set up the system once. After that, a tech-savvy non-developer can make changes with Claude Code or Cursor. For more complex requirements, you need help again.
What about security and deployment?
The agents can also deploy. You have a repository, two branches (development and live), and the agent handles the rest. Sounds technical, but the agent explains every step.
Will CMS systems really disappear?
Probably not completely. But they will change radically. Today’s interfaces — built for humans — will become APIs that agents use. The CMS becomes a database, no longer an interface.
Who is AI-first suitable for?
For companies that:
- Have someone who wants to experiment
- Have no critical dependency on the website
- Need to save costs
- Want to learn the future
What about complex websites and portals?
That will take longer. Large systems with rights-roles concepts, workflows, digital asset management — they’re not built for agents. But even here, the interaction will change.
Conclusion
The truth is probably somewhere in the middle: AI-first will come, but slower than the enthusiasts believe and faster than the skeptics expect.
For companies, this means: Try it out. Take one process, experiment, learn. Don’t switch everything at once, but also don’t wait until everyone else shows the way. How affordable custom software has become with AI agents makes getting started easier than ever.
We’ll talk about this again in three years.
This article is based on a debate between Manuel Zorzi and Michael Kirchberger about the future of CMS systems in the AI era. Watch the full podcast episode on YouTube →