Thoughtful AI integration beats bolted-on Clippy
Like many, I've been exploring how AI can assist my work and that of my colleagues. The theme is consistent: it’s about giving AI the right context so it helps us do our jobs better — not replace us.
I appreciated this piece from Lullabot: Smart AI integration: moving beyond chatbots and quick fixes
There’s a difference between AI that’s thoughtfully integrated and AI that’s just tacked on. Too often, we encounter AI implementations that feel like modern-day Clippy, popping up to “help” us in ways that miss the mark. Most AI implementations feel like afterthoughts...because they are.
This is good mental fuel (and a persuasive piece to share) and I’m looking forward to expanding on a small prototype we built in Drupal (D10.4, ECA 2, and the contrib AI module). With a bit of glue code to expose a PDF’s extracted text as a token and a new “Send to AI” workflow state, editors can kick off assistance when it’s useful — not by default.
In the example workflow, the system:
- fetches the text of an uploaded PDF
- sends it to ChatGPT
- returns a short summary
- proposes a title
But that's just the technical tooling, we still need to make sure the AI has a well document IA to leverage.
See also: Building guidance for your AI#
Once you've integrated AI thoughtfully, the next step is ensuring it produces consistent output aligned with your standards.
For how to extract patterns from your best content and create actionable guidance, see Improving AI chatbots with an editorial handbook from your best content.
For the complete framework connecting integration strategy, pattern extraction, and measurement, see From irresponsible AI to evidence-based editorial guidance.
For how to test whether your guidance actually works (and iterate based on evidence), see From irresponsible AI to evidence-based editorial guidance.