A few weeks ago, a paper landed in my inbox from the Journal of Esthetic and Restorative Dentistry. I almost skimmed past it. I'm glad I didn't.
I read it twice. Then I read our own article again. Then I sat with it for a while, because what I was looking at was something you don't see very often in this industry: two completely independent groups, who had never spoken to each other, arriving at the same conclusion from opposite directions.
In February, we published Why the Future of High-End Dentistry Must Be Inefficient. The argument was strategic. We looked at where AI was heading in implant dentistry and made a case that the elite labs and clinicians would not be the ones to automate the fastest. They would be the ones who understood which parts of the workflow to protect from automation, and why. We called it strategic inefficiency.
In December 2025, Awdaljan and colleagues published a clinical case report in JERD describing a hybrid workflow for an anterior metal-ceramic crown. Their argument was clinical. They weren’t making a philosophical point about the future of the industry. They were trying to solve one of the hardest practical problems in esthetic dentistry: matching a single central incisor on a tooth with a metal post-and-core, which darkens everything underneath it, to a standard that could withstand scrutiny next to its natural neighbour.
They had no idea we existed. We had no idea they were working on this. And yet the workflow they designed reads like a clinical translation of the philosophy described.
What makes this paper remarkable is not the technology, but the decision about where to deploy it.
The researchers used an AI software called Matisse to analyse the patient’s existing tooth colour using a colourimeter, then generate customised staining recipes for three crown zones: incisal, middle, and cervical. The software calculated precise pigment ratios, accounting for the optical properties of each zone and the darkening effect of the underlying metal substructure.
That is impressive engineering, but it is not the interesting part.
The interesting part is what happened next. The clinician took those AI-generated recipes, sat chairside with the patient, and applied the stains intraorally by hand. With a fine brush, under isolation. Making visual comparisons in real time against the adjacent natural tooth, adjusting pressure and layering by feel, working with the kind of micro-level dexterity that no robotic arm currently replicates in a living mouth.
The AI did the colour science. The clinician did everything else.
The final colour difference was 0.5 ΔE₀₀ units. For context, the perceptibility threshold (the point at which 50% of trained observers can detect any difference at all between two objects) sits at 0.8 units. This restoration didn’t just pass. It was, for practical purposes, invisible. And they achieved it the first time, without a remake, without a second appointment, without sending it back to the lab for correction.
Now, a certain kind of technology evangelist would look at that result and say: “Excellent, so let’s automate the application step too. Get the human out of the loop entirely.”
That response misses something fundamental about why this workflow succeeded.
A fully automated version of this case would require fewer steps, cost less, and provide a faster turnaround. It would also have removed the clinician’s ability to see the restoration against the living tissue, under the actual lighting conditions of the patient’s mouth, with the subtle moisture and translucency of natural teeth right there for comparison. It would have removed the feedback loop that only exists when a trained pair of human eyes is making judgments at the point of delivery.
The researchers chose the longer path. The AI generated recipes. A dental technician built the ceramic foundation by hand, layering feldspathic porcelain to create a high-lightness base. The clinician then applied stains chairside. The restoration was returned to a furnace for final glazing, which meant more handoffs, more steps, and more opportunities for human judgment to intervene.
That is, by any efficiency metric, a worse process. It is also the process that produced a sub-threshold colour match in a single attempt on one of the most unforgiving restorations in our field.
In our original article, I wrote about cognitive surplus: the principle that when you offload precision engineering to the machine, you don’t just save the technician time. You free their entire cognitive bandwidth for the things the machine cannot do. Wisdom. Contextual judgment. The instinct, built over years, that something is not quite right even when the numbers say it should be.
This paper presents the principles of clinical practice. The clinician was not burning mental energy trying to determine which pigment ratios would produce the target CIELAB values for each crown zone. Matisse had solved that problem. The clinician was free to focus entirely on application: how the stain was sitting on the ceramic surface, how the colour was reading against the gingival tissue, whether the incisal translucency looked alive or flat. The kind of assessment you can only make when you are in the room, with the patient, under real conditions.
The paper’s own discussion section describes this as demonstrating “the synergy between digital technology and artisanal expertise” and states that the AI “complements” rather than “replaces” traditional methods. The authors, writing from a clinical research perspective with no strategic agenda whatsoever, independently arrived at the same formulation we reached from the opposite direction.
Reading this paper felt oddly familiar because it describes something I do every week in a different context.
When I finalise an Atlantis design (and I have personally signed off on over 10,000 of them now), the process is structurally identical. The software generates a design proposal based on the scan data and the implant position. That proposal is good and mathematically sound. It accounts for the restorative space, the emergence profile, and the abutment geometry.
But it does not know that this particular patient is a heavy bruxer who needs a flatter occlusal scheme. It does not know that the buccal tissue quality is compromised and that the emergence angle needs to be more conservative to avoid recession. It does not know what I know because I have been looking at these cases for over two decades, and because I spoke to the clinician on the phone last Tuesday about their specific concerns.
So I review the software’s draft, and I accept what works. And when I need to, I override it. The technology does the engineering, and I provide the editorial judgment. That is the same workflow this paper describes, operating at a different scale and in a different clinical context, but governed by the same logic.
I wrote in our original article that the lab’s value would shift from creation to curation. This paper shows that the shift is already happening.
There is a concept in science called convergent evolution: when two organisms, evolving in completely separate environments with no contact between them, independently develop the same solution to the same problem. Sharks and dolphins both evolved streamlined bodies and powerful tails, not because one copied the other, but because the physics of moving efficiently through water demands that shape.
Something similar is happening here. The strategic argument for keeping humans in the AI workflow and the clinical evidence for doing so are converging on the same answer, from entirely independent starting points. That is not a coincidence. It is a signal.
The industry conversation about AI in dentistry is still dominated by the automation narrative: faster scans, instant designs, fewer human touchpoints. And for commodity work, that trajectory makes sense. But the clinical evidence is beginning to show that the highest-quality esthetic outcomes are not emerging from the workflows with the fewest steps. They are emerging from the workflows that are intelligent about which steps to automate and which to protect.
The labs and clinicians who understand that distinction will define where this profession goes next. We have always believed this at Nexus. It is encouraging, to put it mildly, to see the peer-reviewed literature arriving at the same place.
The doorman was never the bottleneck. He was the value.
The paper referenced in this article: Awdaljan MW, Alvarado A, Santos E, Roque JC, dos Santos Nunes Reis JM, Rondón LF. “A Hybrid Approach Using AI-Assisted Chairside Staining for an Anterior Metal-Ceramic Crown.” Journal of Esthetic and Restorative Dentistry, 2025. https://doi.org/10.1111/jerd.70079
Read our original article: Why the Future of High-End Dentistry Must Be Inefficient - https://nexus.dental/news/why-the-future-of-high-end-dentistry-must-be-inefficient/
