M.025 Nothing and everything, all at once
On the collapse of the product-process distinction
Three and a half years ago, ChatGPT launched and transformed AI from a topic for the few to a topic for the many.
Since then, AI has changed absolutely nothing. And it has changed everything. At the same time.
This certainly sounds like a hallucination, but it isn’t. Let me explain why.
Nothing has changed
The promise, hype and investments around AI the last 3.5 years have been massive. So has adoption of tools like ChatGPT. Most of us got a new icon or two on our home screens, and many are even paying for access to better versions of the apps hiding behind those icons.
But if we look around and remain honest with ourselves: not much else has changed.
Most organizations look and feel pretty much the same as in 2022. The people, or at least the types of people, are the same. The processes are the same. The structures are mostly the same. And the products or services have developed along the same trajectory as we were on in ’22.
This is true where I work, and it’s true in the majority of other organizations.
If we zoom further out, society as a whole is also pretty similar. The players are mostly the same, the institutions are the same, and so are the larger social systems in which we operate.
Most things are remarkably similar today to what they were 3-4 years ago. Seemingly untouched by AI.
The fact that so little has changed from AI might seem like a mystery in light of the hype and investments. But the explanation is pretty straightforward: it takes time for people, processes and social systems to adapt to new products.
Processes have always followed products. When a new technology application arrives, whether it’s the internet, email or the telephone, most of us first look at the new with (healthy?) scepticism. Then we carefully try it out within our established workflows. Finally, if we are sufficiently convinced, we start to rethink and redesign those workflows around the capabilities of the new. And first then, will the real potential of the new start to reveal itself.
The issue is that most of us are in the early stages of this process. We either look at AI with scepticism, or slap Copilot on top of existing processes designed for a non-AI world. While we scratch our heads, wondering why the revolution hasn't kicked in yet.
In other words, nothing has changed because for most, the adapting and rethinking of our processes around these new capabilities just hasn’t happened yet.
Everything has changed
While nothing has changed on the surface, things are fundamentally different below it. There everything has changed.
Everything in the sense that a range of key assumptions most of us take for granted have shifted. Assumptions like “knowledge is a scarce resource”, “competence is in limited supply”, and “knowledge work requires time, money and effort”, are challenged in area after area as we speak.
One area where this change has gone further and reached the surface is in software development. Where vibe engineering tools like Claude Code allows us to do in minutes or hours what would previously require days or weeks of work.
The efficiency gains of such advances are impressive, but the deeper and more interesting implication of these developments lie elsewhere: that the increased speed and lower costs associate with AI-work allows us to reverse the very direction of the mentioned product-process relationship.
Product follows process?
The rapid building speed with AI implies that we no longer need to adapt our processes to products designed to work across organizations and users. Now, we can increasingly often instead design whatever process we think is best in a given context, and use AI to build and tailor products to that specific process.
Product can suddenly follow process to a much greater extent than ever before.
But more than being a potential, this reversal is here, right now. Let me give you a first-hand example.
At NHH I am in charge of the mandatory bachelor course in strategy. Every semester my 400-450 students have three full-day workshops, where they in each work on a complicated strategy case in groups over six hours. Normally a workshop is led by myself, supported by a team of 1 PhD student and 8 master students that help me supervise the 120 groups.
This spring I had to rethink the set-up, as I had a large external keynote on the day of the first workshop, and I had three student assistants fewer than usual. With the help of my colleague and AI-partner-in-crime Alexander Lundervold, we built a system that would guide and tutor the students through the entire full day workshop. While the professor in the course, me, was occupied elsewhere.
A multi-agent AI system built for one specific workshop-process, in one specific course, at one specific school. Product followed process.
The AI-workshop system is an example of how AI and the product-process reversal allows us to do things we already do differently. Suddenly a digital version of the professor could tutor the students, patiently, six hours straight. Compared to a few minutes per group in a regular workshop.
But the same logic also extends to things we don’t do today at all. When the time, money or effort required to do knowledge work drops towards zero, we can suddenly do all sorts of things that didn’t make any economic sense when the work had to be done by humans.
For example, we can now update our strategic analyses every month, week or even morning, instead of once a year. We can run personalized, interlinked workshops for thousands of people in large companies (which is what we’re doing at OAO, the startup I co-founded). Or we can build a digital version of the NHH rector that interviews 1000 people to get input to the school’s new strategy.
Beyond opening a range of new opportunities, the product-process reversal also can have competitive implications. So far, we mostly discuss how the high speed and low costs of AI makes it easier to imitate existing products and harder to sustain competitive advantages over time. But as more and more companies start to design their own idiosyncratic processes around the new AI capabilities, and then build customer products into those processes, things start to change. Then the one-size-fits-all approach to software will be replaced by processes and technology applications that are more dissimilar across firms. And make any advantages harder to attack in the process.
All because AI now allows all of us to develop the products we need for whatever process we think is best.
Process equals product?
That said, the product-process reversal isn’t the end of it. It’s more the beginning of something even bigger and stranger: a world where the distinction between product and process might collapse entirely.
And that too is in a sense already here. Let me give you another first hand example.
A few months ago, as part of NHHs strategy process, Alexander Lundervold, Lasse Lien and myself ran a full day strategy workshop with NHH’s top 45 leaders. Right before the workshop started, we got the idea to build a tailored piece of software to elevate this particular workshop. So we did. Alexander spun up his coding agents, we refined and added ideas as the participants worked in groups, used the product during the workshop, adapted the workshop process to the new product, adapted the product to the updated workshop process, the process to the product, and so on. Both the tool and the workshop worked like a charm, and we have never used it since.
The process was the product, and the product was the process. The distinction between the two had collapsed.
Weird indeed, but not really special. Because the same trend is also visible elsewhere. At the frontier, labs are experimenting with coding models so fast that when you hit a button on a website, it takes you to a page that didn’t exist until your click. It gets written the moment you press it. Anthropic demonstrated such a system with their retro looking “Imagine with Claude” last year. Google launched a demo of their Gemini Flash model earlier this year to do much the same, to generate software as you consume it. And more recently, Flipbook, a visual browser that generates new pages as pixels while you surf, also push developments in this direction.
But it doesn’t stop there. Developments of world models, like Google DeepMind’s Genie 3, point in the same direction. World models like Genie 3 generate interactive 3D environments from a text prompt, frame by frame, at 24 frames per second. When you navigate the 3D environment, the next frame you see is produced in response to where you move. Continuously. Your process of interacting with this world becomes the product. And the product is the process. A consistent world that is generated and regenerated as you explore it. Tomorrow, the same technology could be fuelling software where screens become pixels generated as you use it.
Different angles, but one common direction. Products, software and even full virtual worlds produced as they are used. The product becomes the process. The process becomes the product. The distinction we have taken for granted dissolves in front of us.
All at once
Where does this leave us? Nothing and everything has indeed changed with AI. But more striking than any one development or example, is the fact that nothing and everything is happening at once. Two very different realities, running in parallel, as we speak.
Most are still stuck in the first dimension, where AI is mostly a new icon on the home screen, that we try to wrap our heads around.
The other dimension is emptier, weirder, and far far speedier. But its still very real. And its open to both visitors and new residents. The only requirement to see it is that you behave as if some key assumptions underlying your work have been falsified by AI. Imagine what you might build if you put all the capable, but currently idle digital knowledge workers in your office to real work. Explore what process you would design if you don’t have to wait for someone else to build a product first.
If you do, you will likely realize that in this new dimension the key constraint is not capacity, as in the other dimension. The key constraints are ideas and imagination. It is knowing what to ask for. It is having problems worth solving and ideas worth pursuing.
But there are similarities too. The humans are the bottleneck also in the new. Just in a different way than before. Where progress is bound by their ideas and good problems, not by their capacity.
The two dimensions are still parallel, but likely won’t stay like that forever. At some point they will collapse into one. Just like the distinction between products and processes already. And then you will see it, whether you like it or not.