#525 The Delivery Gap: Why AI Coding Speed Without Quality Is Becoming a Strategic Business Risk
The Delivery Gap: Why AI Coding Speed Without Quality Is Becoming a Strategic Business Risk
Article by Niels Brabandt EMBA MBA MSc
Artificial intelligence has dramatically altered the economics of software development.
Code can now be generated at unprecedented speed. Teams that once struggled to deliver minimum viable products can suddenly produce prototypes in days. Pull requests increase. Experimentation accelerates. Executive optimism rises.
Yet beneath this technological acceleration, an uncomfortable reality is emerging.
Many organisations are producing significantly more code while generating disproportionately less business value.
In this week’s Leadership Podcast and Videocast, Brenn Hill joins Niels Brabandt to discuss one of the defining technological leadership challenges of our time: the delivery gap between AI-generated software output and production-quality business outcomes.
The issue is not whether artificial intelligence can write code.
It clearly can.
The more important strategic question is whether organisations are actually building the right products, with sufficient quality, at economically rational costs.
According to Brenn Hill, many organisations have misunderstood what success in AI-enabled software development truly looks like. Businesses celebrate output volume while failing to measure delivery effectiveness.
Pull requests rise dramatically. Yet production quality frequently stagnates or deteriorates.
Incidents increase. Technical debt accumulates. Maintenance burdens grow. Compliance concerns intensify. Costs escalate without measurable returns.
The consequence is a dangerous illusion of progress.
As Brenn Hill explains during the discussion, moving faster into a wall does not create competitive advantage. It merely makes impact more painful.
This distinction matters profoundly for decision-makers.
Executives frequently assume that AI-generated software represents an immediate productivity multiplier. In reality, many organisations encounter a very different experience. Teams generate significantly more code, yet no greater proportion reaches production. Worse still, code quality often declines, introducing bugs, operational risk and reputational exposure.
This challenge becomes particularly severe in regulated industries.
Financial services, tax software, healthcare and compliance-sensitive sectors operate within fundamentally different risk profiles compared to lower-risk environments. Organisations must therefore rethink the simplistic assumption that speed automatically outweighs governance.
The future does not belong to organisations moving fastest.
It belongs to organisations moving intelligently.
One of the most valuable frameworks introduced by Brenn Hill is the concept of the Verification Triangle: inputs, outputs and cost.
The first dimension concerns intent clarity. Artificial intelligence enables organisations to prototype at extraordinary speed. This creates a significant strategic advantage, provided organisations distinguish between experimentation and production readiness. A prototype that works on a laptop remains fundamentally different from enterprise-grade software.
The second dimension focuses on outputs. Quality assurance, verification frameworks and delivery gates become increasingly essential in an AI-enabled environment. High-performing organisations understand that trust in software delivery depends not on optimism, but on systematic validation.
The third dimension addresses cost. Many organisations still struggle to calculate the true economics of AI implementation. Infrastructure, human oversight, token consumption, maintenance requirements and downstream operational consequences frequently remain invisible.
Without visibility into cost structures, return on investment becomes impossible to assess.
This creates a leadership challenge rather than merely a technical challenge.
The organisations succeeding with AI in software delivery are not necessarily those embracing automation most aggressively. They are those that invested long before AI in trusted workflows, quality systems and disciplined delivery processes.
Artificial intelligence amplifies capability.
It also amplifies dysfunction.
The implication for leadership is therefore clear: before organisations attempt tenfold acceleration, they must first learn how to achieve trustworthy, measurable and economically valuable delivery.
The real competitive advantage lies not in writing more code.
It lies in closing the delivery gap.
For more leadership insights on AI, software delivery, organisational performance and productive quality, listen to the latest Leadership Podcast and Videocast with Brenn Hill and Niels Brabandt.
Niels Brabandt
---
More on this topic in this week's videocast and podcast with Niels Brabandt: Videocast / Apple Podcasts / Spotify
For the videocast’s and podcast’s transcript, read below this article.
Is excellent leadership important to you?
Let's have a chat: NB@NB-Networks.com
Contact: Niels Brabandt on LinkedIn
Website: www.NB-Networks.biz
Podcast and Videocast Transcript
Niels Brabandt EMBA MBA MSc
Let's say you want code, and of course you might think, "Well, that became a bit easier because we now have AI, don't we?" And here we're going to talk about something which is happening very quickly. It's called the Delivery Gap, and I have an expert on the matter with me here today. Hello and welcome, Bran Hill.
Brenn Hill
Hi, glad to be here, thank you.
Niels Brabandt EMBA MBA MSc
Thank you very much for taking the time. You just wrote a book, and the book is called The Delivery Gap. And of course, when it's about books, I have to ask one question first, because let's face it, it is not easy to write a book. It takes a lot of time. So my question, of course, has to be, when you said, "I'm going to write a book about the Delivery Gap," what was your core motivation in the first place to write a book in today's times?
Brenn Hill
Actually, I did not originally intend to write this book. I was going to write a different book on AI called AI Augmented Dev, because I have a background in technology and writing code. And as I was working with the technology and diving into how it was playing out in organizations and talking to my network, I realized it was the wrong book. The book about how to use AI to write code was the wrong one. Everyone was already writing code. The problem was they weren't getting anything out of it. They were producing all this code, and it wasn't going anywhere, and no one was getting any return on their investment.
Brenn Hill
And that's what I saw in the data and what I was seeing. And so the original book was completely shelved. I had written a lot of it, and I just scrapped most of the content and basically ended up starting again. And that became this book, which I felt was an important story to tell because even with my current company and my network, I was watching this play out, right? Like, how do we make sure that if we're going to spend all this money and time and energy on AI, we get something for it?
Niels Brabandt EMBA MBA MSc
Yeah, absolutely. I mean, let's face it. When we look at the reality at the moment, what is actually up is the pull requests. The review quality, however, is not, let's say, where it should be.
Niels Brabandt EMBA MBA MSc
So how can this be solved? Because most people will tell you, "Look, we can't bring the pull request number down because we're going faster, faster, faster, faster, so that would rather go up." How do you want to make the review quality better without running into massive risk where compliance is never going to say yes to?
Brenn Hill
Yeah, I mean, that's kind of the crux of the book. So first, everyone should read it. But the thing to understand is it's not just about pull requests. It's about the entire lifecycle of the code you're creating, right? So even if you create a bunch of pull requests and let's say you merge them, if you have incidents later and you're busy answering angry phone calls or your services, that's not helping your business.
Brenn Hill
Going faster into a wall just means it hurts more, right? So it's more about acting with certainty and making sure that whatever speed you get is worth having. Is it better to go 10 times faster in the wall or 2 times faster, but with clarity, certainty, and knowing you're building the right thing?
Brenn Hill
And there's actually a couple different ways, a couple different walls people hit. One is they build the wrong thing. They're so overjoyed with the ability of AI to just make stuff that they make tons of stuff, whether they need it or not. And then they have to maintain it.
Brenn Hill
People think that the hard part, the expensive part of code is making the code. But it's really that once it's in production, once people are using it or even not using it, you're also spending your time, your energy, your tokens, whatever, maintaining it. And then even if it is the right code, can you trust it, right? Can you trust that what you've sent to the production actually works the way you want it to? So there's a lot of ways this can all go wrong.
Brenn Hill
And then, of course, there's just cost control. How much did it actually cost you to deliver? And it turns out a lot of organizations can't really measure this either. So they don't even know if they're getting a return on the investment.
Niels Brabandt EMBA MBA MSc
So when you now talk to organizations, because I, of course, prepared for this interview, when I talk to other organizations, they said, "Look, this is just how things are. Either you can do everything we do with human-proved, so we review it with humans, which is too expensive, or we go into a certain risk and we do it automatically somehow, but then the risk goes up, and that is something you have to live with."
Niels Brabandt EMBA MBA MSc
Would you agree on that approach, or do you think that you can go, let's say, low risk but not 100% human-approved because that might be the too expensive approach?
Brenn Hill
You're always taking, even if you want all humans, you're taking some risk, right? So the risk-reward trade-off depends on what you're doing and what your business is, right? So if you're doing tax software and it has to survive audit, or regulated industries, you have a different risk profile than, say, something that makes cat cartoons for YouTube, right? Like, completely extreme, different risk profiles.
Niels Brabandt EMBA MBA MSc
Yeah, that's true. Yeah.
Brenn Hill
And you've got to just be really honest with yourself about what is your risk profile. And then, yeah, but you can blend them, right? So a lot of the book is about this. It's like, what do you need to put in place, whether you're regulated or not?
Brenn Hill
So there are organizations like Stripe that have been very successful shipping software with AI, but they're doing financial software. They're a regulated industry. So it is possible to do. But the thing about these organizations that are successful doing it is they've invested, long before AI, a tremendous amount of energy and money into verification and quality software and setup and workflows so that they can move fast, even with humans, in a trusted way. And now, in the age of AI, it really pays off.
Brenn Hill
But for most organizations, they never invested that hard into that sort of quality control software and processes and so on. And so when they have AI added, they go really fast, and then it explodes. They just don't have everything tight like that. And so a lot of the book is about what do you need to put in place as a company, regardless of risk profile?
Brenn Hill
Obviously, if you're higher risk, you need to do it more so that you can get that speed up, because most organizations, they don't get any speed up at all. In fact, a lot of them go bad. They spend a ton of money on AI tokens. They create three times as many pull requests, no more get into production, so they aren't going faster, and the ones they get into production are buggier. So they spend a lot of money to have a worse outcome.
Brenn Hill
Well, that's not what we want as a business. I would rather go two times as fast, not 10, but have it actually be worth something. And so this is what I work with organizations about, is like, how do you get and start at 2? If you can't get to 2, you're not going to 10, right? So how do you get to 2 first?
Niels Brabandt EMBA MBA MSc
I think that's a great point here. And you talk about something in the book, and of course, anyone should buy and read it. You call it the Verification Triangle. Can you give a bit more detail on the Verification Triangle?
Brenn Hill
Sure. So basically, I give a very simple model. This is more for management on where to look. There's inputs, there's outputs, and there's cost. That's the triangle.
Brenn Hill
So the input is about getting intent clarity. AI makes it very easy to be agile in a way that people have talked about agility for a long time, but it's like, how do you know you're building the right thing? And the thing is, with AI, you can actually iterate and prototype very, very quick. The final output may not be production-worthy. This is the disconnect.
Brenn Hill
People build a prototype, and they think, "Oh, we'll put this into production. It's ready because it looks good or it works on their computer." It's not ready. But you can quickly test to see if it's at least the right thing, right? So use that, leverage that to iterate really fast to find the right thing to build the correct way, and then invest in building that right. And that's your output. And there's a bunch of ways that I talk about, like quality gates and things like that. Measure the quality of your output, right? So that's a second gate.
Brenn Hill
And then all of that has a cost, a fully loaded cost: tokens, human time, infrastructure that you've put into it. And a lot of times, companies don't even know how to measure this, and so they're not able to measure ROI. So you want to have a project. What is the input? Do you know you're building the right project? You finally get the output. How much did it cost you, and where did those costs go?
Brenn Hill
And so if you're an engineering leader like myself, this is what you need to know. And so I give some examples of how you calculate this and dive into it, and then you know where your energy, time, money is going, and whether you're successful or not. And then if you need to go to your CEO or your board or whatever, you've got all the data because you know what data to get.
Niels Brabandt EMBA MBA MSc
Yeah, absolutely. And of course, you know what you're talking about because I thoroughly background-check every single person that gets on this podcast. So you have a real-world career here. You're American, live in Berlin, you worked internationally, you worked in Asia, all over the world. So you really are in these projects.
Niels Brabandt EMBA MBA MSc
And now some people might wonder, "Okay, look, I'm going to buy the book. I'm going to read the book." And then some people might think, "I think we still need help here." And I think you can offer help here.
Niels Brabandt EMBA MBA MSc
So of course, the final question of the interview is, how can I get people in touch with you when they say, "I think Bran could be a great contribution, either as a keynote speaker or a coach or a trainer or someone working on our projects"?
Brenn Hill
Yeah, just reach out on LinkedIn. I'm very easy to reach. Or send my email or any other ways. Like, I have a website on Substack. Any of those ways will work, and I'm happy to hear from people.
Niels Brabandt EMBA MBA MSc
Excellent. So at the end of this podcast, at the end of the video cast, we now know the Delivery Gap is the one you need to close, and Bran is the person to talk to. So at the end of this podcast and video cast, there's only one thing left for me to say. Bran, thank you very much for your time.
Brenn Hill
Thank you for having me.