Oh look, I got peer reviewed and published.

Took a couple of decades to get fed up of noticing the same gap.

I wrote a paper. It's called "The Design Integrator Role: Governing AI from the Inside," and it just came out in Touchpoint, the SDN journal. I'm not reproducing it here, that's not mine to give away, but I want to tell you why I wrote it. The why is more useful than the abstract anyway.

The gap isn't new. The vocabulary is.

For most of my career years across service design and digital experience, I could feel exactly where a service was going to fail a person, well before it failed them. I could raise it or could design around it where I had room. I just could never stop anything. That authority lived with risk, with compliance, with the model owners. Never with whoever's job it actually was to ask if the human understood what was happening to them.

Not a complaint about any one employer. It's what happens to design teams generally, inside organizations that don't quite know what to do with them.

I took an AI executive course at Rotman earlier this year to get an educated viewpoint on AI and systems with a business growth lens. I got what I wanted, mostly the vocabulary of model risk, deployment gates, halt criteria, a whole grammar that already exists for financial risk, reputational risk, model drift. The gap, glaringly for me at least, was whether the human on the other end of the agentic/generative/insert-AI-type-here was actually served. Once I had the words, the paper took a few months. Living inside the gap without them took years.

What's actually in it

Nobody formally owns what I call the human outcome layer in AI-mediated on-boarding: whether the customer understands what's happening, whether they have a genuine choice, whether they have a real way out. Compliance checks whether disclosure exists, not whether it landed. UX can ship a beautifully designed misunderstanding right on schedule. Neither one owns this layer, and the org chart doesn't draw a box for it either.

The proposal is a role, not a value statement. The Design Integrator: formal authority to test an AI-mediated flow against comprehension, consent and exit criteria before it deploys, and formal authority to halt deployment when those criteria aren't met. Sitting at the table the same way model risk already sits there. Not advisory. An actual seat.

What I want practitioners to actually take from it

Not "AI ethics matters." You already know that, and it doesn't ship anything.

Name the layer. "Human-centred AI" is decoration until someone can point to who owns comprehension, who owns genuine choice, who owns exit. If nobody in your org can answer that, you have the same empty chair I'm describing.

Test before deploy, not after complaint. Every other risk category gets pre-deployment testing. The human outcome layer usually gets a post-launch complaint line instead, wearing the same clothes as testing without doing the job.

Authority has to include the power to say no. A role that can flag concerns but not halt a launch is advisory theatre. I've sat through advisory theatre before, in rooms that had nothing to do with AI. It performs the function, it doesn't do the job.

If you want the full argument, the governance model, the halt criteria, the case for where the role sits, it's in Touchpoint. You can read the abstract and intro and get to the full paper here.

I spent years feeling the gap before I had the vocabulary for it. If you're a practitioner who's felt the same thing and didn't have the words either, this is for you.