What happens when teams expect new technology to solve problems they have not properly defined in the first place? In this episode, I explore why so many AI projects struggle to deliver consistent value in day-to-day operations. The issue is rarely the speed or capability of the technology itself.
Drawing on ideas from specification-driven development in software, I connect those lessons back to production environments, where the same pattern can play out in different ways. Whether it is signage artwork missing key production details, packaging files arriving with conflicting instructions, or teams expecting automation to resolve ambiguity, the result is often the same: AI can accelerate the work, but it can also accelerate the chaos.
The episode includes selected excerpts from my conversation with Ivo Dusch, Global Business Development Manager at Canon Production Printing, about Height IQ — an AI-powered application designed to simplify a specific part of the elevated print workflow. Giving a practical example of what happens when AI is applied to a tightly defined problem rather than treated as a catch-all solution.
I examine why human review still matters, how bottlenecks shift when one task becomes faster, and why better outcomes usually start with better definitions.
It is a strong listen for anyone trying to make sense of how to implement AI in real production settings — and for anyone interested in how clearer thinking upstream can lead to better workflow downstream, in any industry.
You might also enjoy this blog about technology?
Stephanie Gaddin 0:01
This is Bits and Bots for Business.
Stephanie Gaddin 0:07
The podcast about technology, systems, human ingenuity and automation, from apps to AI, e-commerce to emails, we talk about it all. I’m Stephanie Gaddin, global citizen, technologist and curious human being.
Stephanie Gaddin 0:23
Welcome to episode two of Bits and Bots for Business. We’re talking AI today.
Stephanie Gaddin 0:29
So I’ll start with the question. What happens when we expect AI to fix problems that we haven’t clearly defined, and what good is going fast and accelerating if we’re unclear on our direction.
Ivo Dusch 0:43
So we thought AI is the next big thing. Let’s see what we can do with AI.
Stephanie Gaddin 0:50
That was Ivo Dusch, the Global Business Development Manager for Canon Production Printing based out of Venlo in the Netherlands.
Stephanie Gaddin 1:00
We’ll hear more from Ivo a little later in this episode.
Stephanie Gaddin 1:05
But first, I want to discuss the question I asked.
Stephanie Gaddin 1:09
What happens when we expect AI to fix problems, the ones that we haven’t yet clearly defined. Across industries, from enterprise software into print, signage, packaging. We’ve seen the same pattern emerge through 2025 and heading into the beginning of 2026. AI moves fast. The tools are impressive. Capabilities have expanded, but people and many teams are still struggling to see a consistent, reliable improvement in
Stephanie Gaddin 1:41
their day to day.
Stephanie Gaddin 1:42
And over the past year, I’ve been watching some organisations across our industry and in adjacent manufacturing sectors where they’ve experienced and been experimenting with AI inside their production workflows. And while the tools and the industries differ, there is there is a pattern that keeps showing up, and that is that the speed of AI is not the problem.
Stephanie Gaddin 2:06
The real challenge, and where people fail and where AI integration usually falls over is that teams try to add an AI something into their existing way of working without improving the thinking that feeds that process. I had another conversation recently with a technical software architect, that I’ve been collaborating with, and he has put it simply as “AI does not get stuck on the code, it gets stuck on the humans”.
Stephanie Gaddin 2:40
And that line has stayed with me, because I think it applies just as strongly in the world of manufacturing and print production as it does to large enterprise IT. In almost every sector, whether we’re looking at finance, retail, government, manufacturing and even printing. Leaders want the same thing, a faster delivery with better quality, a clearer communication and fewer costly mistakes. And AI was meant to help with all of that, but in practical terms, most of the rollouts we saw last year, and some of them that are still trying to go, have failed and stalled. And it’s not because the AI model can’t do something. It’s because the team or the individuals treat the AI as if it was a shortcut, and processes are not built for that, and unclear briefs still make their way downstream into the workflow.
Stephanie Gaddin 3:37
Now if you’ve ever received a half-baked client brief, you will recognise this problem immediately. When the input is vague, the output cannot be anything except unpredictable. And AI is no different.
Stephanie Gaddin 3:54
One of the most effective uses of AI that I have seen in the printing industry didn’t start with a sweeping promise of automation of an entire process. It started with one very specific question, “Where exactly is the real bottleneck in this process?”
Ivo Dusch 4:17
We do elevated print already for quite some time, and what we learned from it is that our customers really need skills, Photoshop skills, to get their 2D image, all the way up to a sufficient height map. So we thought, AI is the next big thing, let’s see what we can do with AI.
Stephanie Gaddin 4:42
Ivo is talking about Height IQ, a free AI powered application that Canon released that simplifies the creation of the digital height maps that are required for elevated printing on their Arizona family of flatbed printers.
Stephanie Gaddin 4:59
What stands out for me is not just the time saving that this application creates for graphic designers and printers, but it’s the restraint. Canon didn’t try to automate the creativity. They didn’t try and replace designers. They focused on one specific, clearly defined task that was slowing down everything else.
Ivo Dusch 5:27
We showed on FESPA, that’s a big technology trade show for the print industry, for large format graphics, here in Europe. We showed a technology study around AI, and that was, let’s say, the front runner of Height IQ. And we gain lot of traction around it. So we thought, hey, there is something.
Ivo Dusch 5:51
And of course, AI, as mentioned, is a buzzword, in also the printing industry. And combinated with elevated print, yeah, that that was really appealing. And so we thought, okay, let’s move from that technology study and take it to the next level, and we start developing the tool. And yeah, in September this year, 25 we were almost ready with with Height IQ, we had to do some check marks.
Ivo Dusch 6:20
And indeed, it really showed what AI is capable of, and especially when it comes to that specific thing in the workflow you mentioned right to create that actual height map. And that saves, on average, about, yeah, 10 hours, depending, of course, on the skill level of Photoshop and the type of image that you use.
Stephanie Gaddin 6:45
So Canon saw a very particular and specific problem, and they found a way to fix it with AI. And this is exactly what we see in the world of specification driven development in software, AI can deliver the most value when the problem is well understood, very tightly scoped and clearly articulated.
Stephanie Gaddin 7:08
So specification development is not a tool heavy technical way, if you think about it, it’s more about a mindset. When you strip the concept back to its essentials, it’s really just a modern way of saying we need to be clear about what we’re trying to achieve before we ask AI to help us achieve it. And I think that idea translates perfectly across into print and signage workflows as much as it does in the software world.
Stephanie Gaddin 7:35
For example, a signage company receives artwork and that doesn’t have any colour profiles, scale, information or substrate, production cannot go ahead if a packaging job arrives with contradictory die-line notes and the client insists that the AI should fix it, the box can’t be made. A marketing team requests a quick rebrand, but they can’t articulate the constraints, approvals or any mandatory elements in the brand. For all of these cases, the outcome is not slow because of the tool. It’s slow because the thinking has not finished before the work begins.
Stephanie Gaddin 8:13
AI speeds up work that is well defined, and it amplifies chaos when the work is not defined.
Stephanie Gaddin 8:22
The second thing that comes up often in my work with people on AI is the misconception that AI removes the need for human judgement, or replaces human judgement, and can be used that way. In practice and in my observations, the opposite actually tends to be true.
Ivo Dusch 8:44
But then, of course, in Photoshop, you need to play a little bit and and there is a human still needed to state, okay, this is good enough, or I still need to do some tweaking and things around it. So especially with the visually and then impaired and for the blind, yeah, there. It’s that specific that it’s really depending on the skill level of the designer.
Stephanie Gaddin 9:13
That distinction matters, because AI accelerates execution. However, humans remain responsible for judgement, quality and intent, and once that acceleration kicks in, something else becomes very, very visible. When you’ve got AI doing the repetitive heavy lifting, the bottlenecks in your process don’t disappear. They shif, somewhere else. In software, they would shift upstream into specification and QA planning and downstream into the testing and validation, which become the new bottlenecks. In print I see the same thing happening. In our example, speaking with evil. Once height map creation has become faster, the pressure now shifts to design, clarity, creative intent, getting approvals for the work and the payment and print validation. So AI is not removing work it’s not taking work away. It’s shifting where the work is required, and from a workflow and flow efficiency perspective, this is actually a very good problem to have, because it shows you exactly where to invest next.
Ivo Dusch 10:33
The printing part still needs some time to print elevated because you can imagine that, just a flat print is less time consuming than putting all the ink drops to each other so. Exactly. So hopefully we see that happen, you know, and because that really means that that elevated print is going to a broader audience.
Stephanie Gaddin 11:05
So this is the moment that a lot of organisations and teams miss. They will celebrate the speed gain from AI or the tool that they’re using, but they don’t adjust the rest of the workflow to support this. The teams that succeed when they use AI are doing something different. They take the specification part of the process very seriously.
Stephanie Gaddin 11:30
They make quality of specification important. They embed that clarity and seeking that clarity upstream, they raise the AI literacy of the whole team. And they measure flow, not just tasks completed. And more importantly, they keep humans firmly in the loop of the entire workflow. Because they don’t take AI output on faith it is always reviewed, always refined and validated every single time. And that’s not a limitation, it’s an operating model.
Stephanie Gaddin 12:05
So why choose this operating model? Well, there’s a very important reason that human review will remain essential. It’s because AI doesn’t just speed things up, sometimes it invents things.
Ivo Dusch 12:22
But then the weird thing of AI comes in, and that is that AI can sometimes add things that are not there. You know, that is a little bit the trick of AI, so, and that is what we call ghosting. So when you have your full detail button on, then you can see some things in the picture that doesn’t add to quality and brings the quality down even. And therefore we need those, those four types of tiling. And then the human eye can select, “okay, this is something, this is the one I need.”
Stephanie Gaddin 12:58
So this is why clarity has to come before AI can be used, whether you’re writing software, designing signage, preparing files for elevated printing, or doing printing on bags or packaging. The AI that you’re using needs boundaries. It needs context, and it has to understand your intent.
Stephanie Gaddin 13:21
Specification driven work is not about trusting AI, more or less. It’s about trusting your processes and defining your processes that you can trust them. Because when that happens, AI stops being a novelty, and it becomes something that is predictable, a valuable part of your modern production factory floor. So before we ask AI to speed things up, we need to make sure we actually know what we are asking it to do, because acceleration without a proper direction rarely ends well for anyone,
Ivo Dusch 13:56
I think for many companies, but also for Canon, it’s a trial and error, because we are still in the in the early days of AI, how weird that it sounds.
Stephanie Gaddin 14:07
We are figuring out where it works.
Ivo Dusch 14:09
Yeah, exactly, exactly. And that is something really and, you know, we are part of the bigger Canon, Inc, organisation. And, yeah, I think AI is here to stay, obviously, and it will evolve even more and faster, and it will become better, and also it will take a giant leap, I think, forward. But yeah, it’s basically up to us. But also, really listen to our customers and designers. “What do they need?” You know.
Stephanie Gaddin 14:49
So taking from what Ivo said there, when we talk about keeping humans in the frame of AI, it’s not just about the oversight or approvals on the work or shifting bottlenecks. It also means keeping people involved in the definition and the specification pre-work, which happens earlier in the process and, even before that.
Stephanie Gaddin 15:12
It is worth pausing to ask a simple question, “Is this something that is needed, even if it’s not yet identified?”, because that’s how canon got to the Height IQ app.
[Music] 15:26
This is Bits and Bots for Business. Thank you so much for listening to episode two. You can follow us anywhere that you get your podcasts and sign up for our newsletter at rocking rose dot technology.
[Music] 15:41
[ Music to end ]