Editorial May 13, 2026

The traditional CMS is dead. Claude is the new editor.

Strong claim. We'll back it up. The shape of the CMS that won the 2010s — a database, an admin UI bolted on top, a plugin marketplace bolted on top of that — is finished. Not because anyone hates WordPress (we don't), but because the workflow it optimizes for is no longer the workflow content teams actually want.

The job of the CMS used to be storing content. The job of the CMS now is brokering a conversation between a human editor and an AI agent. Those are different problems.

The shift

What changed

Three things, all in the last 18 months. First, Claude got good enough at reading a brand voice guide and respecting it that an agent can draft a page that doesn't sound like a chatbot. Second, the tool-calling story matured to the point that an LLM can propose structured changes — not just emit text — and a host application can apply them safely. Third, the cost dropped. Gemini Flash answers a typical visitor question for less than a tenth of a cent. That's a different economic regime than ChatGPT-4 in 2023.

Together those three shifts mean the bottleneck in publishing a content site is no longer "writing the page." It's "reviewing what the agent wrote." Which means the CMS that wins is the one that makes review fast — diff-based, audit-logged, one-click apply or reject. Not the one with the prettiest WYSIWYG.

The diagnosis

Why the database is the problem

If your content lives in MySQL, an AI agent can't read it without API plumbing you have to build. It can't diff a proposed change against the current version without a custom workflow. It can't roll back without a database migration. The agent ends up bolted on through plugins — and now you're back in the plugin-CVE economy, except the plugins are AI plugins and they have access to your content database.

Filesystem storage solves all of that with no new code. The agent reads markdown files. It proposes edits as text diffs. You review in the same chat where you asked for the change. The "audit log" is git log. The "rollback" is git revert. The "migration to a new environment" is rsync. None of that requires a vendor.

The honest part

What Claude actually does well

Not everything. We'll be honest about the limits. Claude is a senior editor, not a magic wand — it drafts, it rewrites, it summarizes, it answers questions about pages you've already written. It does not invent product strategy. It does not decide what your site is for. It will make confident factual errors if you let it answer questions you haven't given it source material for.

What it does well, when wired up properly, is the 80% of content work that's tedious for humans: rewriting a paragraph in a different tone, generating meta descriptions, drafting alt text, summarizing a long article into a teaser, catching inconsistent terminology across a site. The human still approves every write. That's the workflow. It's faster than doing the work by hand and slower than letting the AI run unsupervised — which is exactly where the quality bar lives.

The replacement

The new shape

The CMS of the next decade has four properties the old shape doesn't:

Content the agent can read and write directly. No API tax. No plugin middleware.
Staging-first writes with reviewable diffs. Human approval is the safety mechanism, not the bottleneck.
Multi-provider AI with the customer's key. Vendor lock-in on the LLM is yesterday's mistake repeated.
Customization in real code. Razor, C#, CSS — not a proprietary designer tool that exports nothing useful.

InkCMS is one implementation of that shape. There will be others. The point isn't to convince you to switch from WordPress to us tomorrow — most sites don't need to switch this year. The point is that when you do reach for a new CMS, the questions you should ask have changed. Not "how big is the plugin marketplace" but "can Claude read my brand guide and respect it." Not "what's the WYSIWYG like" but "what does the approval diff look like." Not "what's the monthly price" but "do I bring my own AI key or pay your markup."

The caveats

Where we'd push back on ourselves

The agent-native workflow is not the right shape for every team. If you have fifteen editors co-authoring in real time, Sanity's collaborative model is still better than ours. If your content team lives inside a marketing-cloud suite and never wants to see git, a hosted SaaS is still easier to onboard. If you're shipping a global news site with millions of pages, the database starts to earn its keep again.

But for the broad middle — content-heavy sites, small editorial teams, one or two writers and an AI — the traditional CMS is dead. The replacement is a filesystem and a senior-editor agent. We built one. Here's how it stacks up against the CMSs you're already paying for.