Why we're deliberately (still) totally not an AI-first Headless CMS

TLDR

• Every CMS we're seeing seems to have gone into AI-first mode. We still haven't.

• We're not "anti-AI" by any means, we're very AI-friendly: we use Claude Code internally, our users build dope things with Claude Skills and Cowork automations, and we have actual AI tooling available for those who want to use it. The key word being want to use.

• We just won't ship features that sit unused and quietly route your content to an LLM you never opted into just to add a checkbox on feature bloat. Are we leaving money on the table with that decision? Most likely.

• As a European company we still have some concerns about what we push through LLMs ourselves, so we're not really comfortable acting all willy-nilly with your project's data.

Wait, are you not building AI features?

If you've opened almost any CMS dashboard recently, you've probably seen an AI writing assistant, an AI SEO optimizer, an AI image tagger, an AI-agent mode, an AI workflow builder, an AI dishwasher manual, and if you're especially lucky, a full-screen modal asking if you'd like to "superdupercharge your content strategy to the moon for ultra enterprise scale with AI to unlock generational efficiency and velocity."

You clicked past all of it and went back to work.

We know that because that's what you've been telling us.

But some of you on the other side of the discussion are also curious why we aren't shipping more AI features.

The TLDR of that is we've been watching, thinking, and deliberately not shipping native features for things that aren't ready, or aren't applicable enough to the vast majority of you.

On a high-level, DatoCMS and every Headless CMS is already AI-friendly by default. Our APIs and CLI are structured, clean, and straightforward enough that plugging them into any AI workflow is smooth. You're not fighting the CMS to make it work with your tooling. The content is already there, already structured, already query-able. We just haven't found the need to add an AI layer on top, and we decided not to drop noise so as to not get in the way of your work.

To be clear, we're not anti-AI. We just think there's a meaningful and practical difference between AI that's genuinely going to be used by most of you in your daily workflows, and AI that's been slapped into the CMS because the "industry demands it". We'd rather ship the former, even if it means shipping less.

That being said we're also not sitting on our hands either. We've been pushing out AI-related things, just not as core features you're forced to interact with. Everything we've done on that front is opt-in, for the use cases where it actually makes sense for you where you're free to use it if you want, and not forced to ignore it and work around it because a few others wanted it.

What the industry shipped vs. what users actually think

Honestly though, AI in content tools is objectively a great idea. There are tons of great things you can do here, and we don't doubt that. The problem isn't the vision, it's the execution speed and the incentive to ship something rather than ship something good.

To go beyond our bias of our own opinions and users, we asked for a few insights in a few developer and marketing communities on what they actually thought of the AI features in the CMS they use, and here's some of of that.

...just really noisy but not useful...

...most of it boils down to generating text and I don't always use the CMS to create or write, I bring it from other tools and then edit...

...even for features that make a lot of sense such as polishing things in SEO terms, you'd get mostly a wall of text instead of having it perhaps a bit more integrated into the platform...

...all the great ideas are there, there is a V12 engine underneath, but it's being driven in the backyard...

That's a reflection of what we've been trying to avoid.

So what have we actually shipped?

Nothing native in the CMS that impacts ALL users (I mean, aside from arguably the most useful native AI feature right now for me which is an emoji recommendation picker. I'm at peace with this 💁‍♂️)

The philosophy behind everything this is that our AI features should be opt-in. None of it is baked into your core experience. None of it turns on without you going looking for it.

We're far from "anti-AI"

This is worth being clear about, because "we haven't shipped a dozen AI features" can read as us just being petty haters. It's not meant to be.

We released an MCP server that lets you automate the heck out of your project to your heart's content. Rather than dumping 150+ API endpoints at an LLM and hoping for the best, we designed it with a layered approach that actually fits how models reason. It's there if you want to build AI-powered workflows on top of your content. Completely ignorable if you don't care about it.

We've put some solid elbow grease and real work into making our documentation AI-friendly, because you increasingly use LLMs to help you build things, and we'd rather that experience be genuinely useful for you than hallucination-prone. You can even go to any docs page and generate an .md or open it in GPT/Claude to start with instant context.

And we have our AI translations plugin for teams managing multilingual content. One job. Does it well. Optional. Don't install it if you don't need it.

We've taken a conservative and pragmatic approach to AI-enabled features, and we'll keep updating our page as we add more. We want to enable you to use AI well, if you want to use it. We don't want to make that decision for you.

And, as a matter of fact, we're constantly thinking of better and new AI features to implement, but we're making an explicit differentiation on what is AI for developer workflows (MCP, schema related things, CLI workflows, etc.), vs. what is "shiny" to add in like an agentic modal that mostly seems to go ignored.

If you have any other specific AI use case that you really wish was baked into the CMS for your project, our plugins make that a lot more achievable than you might think. Build one for your own project and make it public or keep it private, or reach out to us because we'd love to collab on making your vision happen and potentially find a way to ship a slightly more agnostic version of its' capability for the broader community if that's relevant to more users. The CMS doesn't need to make one AI bet for everyone when you can build exactly the thing you actually need.

Aside from features that face you, there's also internal considerations and wider discussions on the topic.

We use tools like Claude Code internally. We've watched our users and customers build genuinely cool workflows using Claude Skills and Cowork automations. The technology works. And it works well WITH DatoCMS. It's just that it works best when the person using it has chosen to reach for it, not when it's been decided for them.

There's also something we don't talk about enough as a European company: most AI features in SaaS products mean your content is being routed to LLMs with data residency and security policies that range from "non-EU" to "Trust me Bro."

For a lot of our customers, especially enterprise ones dealing with sensitive or regulated content, that's not a theoretical concern. We're not going to quietly make that call on your behalf by baking AI processing into core functionality.

The question we keep asking is whether someone would actually use a feature without being prompted to? Does it realllllly feel like a part of the product, or like something that got sloppily slapped on? And can we be genuinely transparent about what happens to your data when it runs?

When it does, and when we can build it right, we will, and we'll keep you in the loop on things as they develop.