How can you actually use AI in DevTools content? With Victor Coisne from Strapi
Strapi has about 2,000 blog posts and a seven-step AI workflow to research, draft, fact-check, interlink and maintain them — but Victor Coisne still keeps a human in every step because developer trust is easier to lose than gain.
Victor Coisne is VP of Marketing at Strapi, the headless CMS company. He walks through the actual systems his team uses to create new content, refresh a large archive, find unanswered questions and distribute useful material across channels.
This is a practical conversation about where AI content works today, where it still fails, how Strapi grounds its output in trusted documentation, and why exponentially more content makes human expertise and curation more valuable rather than less.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
How are you using AI to create content for developers without losing their trust?
We are careful because developer trust is hard to gain and easy to lose. AI is inside the workflow, but a human remains involved at every step. Quality depends on the context going in: audience personas, company positioning, writing guidelines and the research or insight that cannot sensibly be outsourced. If you leave the important thinking to the model, garbage in still produces garbage out.
What are the three AI content workflows at Strapi?
The first creates net-new, mostly non-branded content from topics and search demand. The second maintains an archive of roughly 2,000 posts, distinguishing small updates from pieces that need a complete rewrite and at minimum adding an outdated-content warning. The third creates structured integration and feature pages, where a reader mainly needs to establish whether two technologies work together rather than follow a long tutorial.
Which kinds of article can AI write well today, and which still need a developer?
Comparisons, lists such as the top editors for a year, plugin roundups and structured integration pages are good fits. A real tutorial that combines Strapi, a front end, search or video tooling is not. For those we first have a developer advocate or outside creator build working code in a public repository, then create the tutorial from the project. I have not seen anyone automate that whole process well yet.
What does the seven-step article workflow look like?
It starts with a human-defined assignment, then research into the topic and ranking pages, outline generation, writing, fact-checking and code review, internal linking, and finally the metadata needed for search and generative engines. Reusable artifacts describing our audience, company and editorial rules feed every run, so the system does not need to relearn the same context each time.
How do you fact-check AI-generated technical content?
We call Kapa's API from the workflow. Kapa is grounded in Strapi's own documentation and resources and is instructed not to answer when it does not know, so we use a model we already trust from the support experience to check the work produced by other models. Its unanswered questions also reveal gaps in our source material, giving the content team a list of things it should explain next.
How does social listening feed the content system?
Octolens monitors what people say about Strapi and related subjects, then produces a weekly recap with citations. Product and engineering see the highlights and lowlights as well as marketing. Those conversations reveal topics the community is asking about, while a shared Slack channel lets the team claim relevant questions with a check mark. The aim is to help, not to insert a promotional reply every time somebody names a competitor.
What happens when a community question receives no human answer?
On our forum or Discord we can bring in Kapa rather than leave it unanswered. We prefer community collaboration, but a grounded AI response provides coverage when no human gets there. That matters beyond the original thread because questions and answers are indexed and may later appear in Google or an AI assistant.
How should a DevTools company decide whether to invest this much in AI content?
It comes down to channel-market fit. Strapi has a relatively low contract value, developers respond well to content, and we sell a content-management system, so a low-CAC content channel makes strategic sense and also informs features we can build into the product. A company with a different contract value or buying motion should not copy the investment without checking that the economics fit.
Does AI make translating developer content an obvious next step?
The economics are changing quickly. Internationalization used to require expensive translation tooling and people; AI is shrinking that cost. Tutorials may be especially suitable because the code is a common anchor that does not need translation. You should still look at where your community already exists, but localization is moving toward becoming a much easier decision.
If AI creates more and more content, what becomes valuable?
Volume is rising exponentially while average quality falls, so opinionated subject-matter experts, trusted creators and human curation matter more. A person can read widely, select the few things worth attention, credit the original creators, and then use an LLM to structure that work. Generation is only half the system too: AI can repackage strong source material for LinkedIn carousels, X threads, short video and other channels, but the source still needs genuine value.
Links & references
This episode is brought to you by WorkOS.