When we sit down with leaders who want to “do something with AI,” the conversation almost always starts in the same place: Which tool should we buy? Where do we run a pilot? Which team goes first? These are reasonable questions. They are also the wrong place to start your journey with AI.
The better starting point is a thought experiment we keep coming back to. Imagine you had infinite resources, unlimited people. What would your business look like? Not a slightly faster version of today. If you could wish for anything, what would be amazing? That is the business to design for, the one nothing about your current organisational chart, tooling, or headcount has any say over.
Picture that company clearly. Then walk back down the hill to where you stand now and ask what it would take to close the gap. That descent, from an ambitious end-state to a concrete next step, is, in our experience, where software leaders create a solid foundation for a step-by-step implementation of AI.
The vision comes first, then the road back
The “infinite resources” question works, because it strips away the assumption that the way you operate today is the way work must be done.
Think about how a knowledge-based company is structured. Marketing, sales, operations, finance, each is essentially a pipeline of work that humans coordinate. Layers of management exist on top of those pipelines largely because it has historically been impossible for a small number of people to direct all that activity directly.
But what if it weren’t impossible? What if a marketing pipeline, a sales pipeline, an operations pipeline could each run as a set of mostly autonomous, agent-driven processes, with a highly skilled human specialist responsible for outcomes and a dashboard showing what’s happening? That is no longer science fiction. And once you can imagine the end-state, you can reason backwards about which parts of it are reachable this quarter, which need new skills, and which need a genuinely different way of working.
This is the move most transformation efforts skip. They optimise the existing process by 10% or slap a chatbot on their website and call it AI adoption. The opportunity is to ask the bigger question first, then decide how far down that road you can sensibly travel right now.
What it looks like when someone actually does it
We recently sat down one-to-one with Dmitry Shapiro, founder of MindStudio, a venture-backed AI startup that has raised tens of millions from VC funds and angel investors. He described how his own company is built, a useful small-scale picture of an end-state.
MindStudio is roughly ten people. The entire product, front end, back end, infrastructure, internal tools, is built by two engineers working alongside coding agents, with the overwhelming majority shipped by a single founding developer. The company employs more content producers than engineers, on the deliberate logic that in an attention economy, building the product is only half the job; being seen is the other half. Customer support, frequently praised by users, runs on one person augmented by agents watching the inbox and the support tooling, with documentation generated automatically as features ship.
No Jira. No elaborate project-management apparatus. No quarterly roadmap ritual. Communication is a thin layer of Slack and email, a couple of short standups a week, documents in Notion or Google Docs, accounting in QuickBooks. The stack is deliberately light, because the heavy lifting has moved into automated processes rather than into coordination overhead.
A ten-person, AI-native startup is not a thirty-year-old company with established procedures and real customers to protect. But it is a clear image of what “designed from infinite (agentic) resources” can look like in practice, and an image is what we believe a mature organisation needs in its head before it starts the journey.
Build the stack, manager by manager
So how does a company that wasn’t born this way actually move? We think about it as a ladder with three steps, and they map to the model we use with clients: Apply, Reshape, Innovate. Each step builds on the one below it..
Step one: Apply – get your people driving agents. As quickly as possible, engineers and the people around them need to be fluent with agentic tools as the default way they work. In practice that means using all four modes of working with AI: delegating tasks to it, asking it for answers, collaborating with it, and exploring what more it could do for their function. The biggest names are already explicit about it. Google says around 75% of its new code is AI-generated. And Nvidia’s Jensen Huang goes furthest, saying he wants his engineers to spend “exactly zero percent” of their time writing code and all of it solving problems. Blunt as those targets are, they work because habits are hard to shift and a clear number gives people something to move toward. This is the Apply phase, putting AI to work inside the way you already operate, where it’s easy and obvious. Most companies stop here, if they get that far.
Step two: Reshape – get every manager to build a better stack for their team. This is the step many companies miss when having no clear end end-state picture in mind: New processes. The shift looks like this: today, a detected bug drops into a ticket and flows down a slow, linear path of triage meetings and spreadsheets. In a Reshaped stack, the moment a bug is detected, parallel agents enrich it, trace it, and propose a solution, and an engineer simply says “yes, ship that” or “give me three more options.” The meeting, the program manager, the spreadsheet can all still exist; they just stop being the bottleneck. We keep reaching for software development examples because they are the easiest to picture, but the same logic applies to every pipeline in the business: product discovery, marketing, sales, customer service, operations, finance. Each is a chain of work waiting to be reshaped the same way. Every manager starts in the same place, with the core exercise we describe below: imagine your team didn’t exist and ask how much of its work could run agentically. The answer tells you exactly what to reshape, and every manager who does it for their team compounds the gain. At this stage, transformation starts to take place automatically.
Step three: Innovate – do things you could never do before. This is where the infinite-resources vision pays off most. Consider what a good manager would wish for if a genie offered it. A sales leader might wish to continuously enrich every contact, watch what they’re doing, and show up to help at exactly the right moment, work that would be absurd to attempt manually across thousands of relationships. Dmitry Shapiro (MindStudio CEO) runs precisely this kind of system for himself. A set of agents runs automatically in the background on a schedule, several times a day, with no one pressing go. Some enrich his CRM straight from his call recordings and calendar, so every contact stays current on its own. Others work his network on X and LinkedIn: the system spots the people who matter to him, works out whose voices those people actually listen to, and surfaces real openings to engage, along with a filtered feed of what they are posting so he never has to scroll. A separate job looks at the people already in his network and asks one question, how can I help this person, purely to be useful, with nothing to sell.
He uses this intelligence to decide when and how to manually act, but acting takes a fraction of the time and the signal is far better. The running cost is a few dollars a day. The point is that it has become easy to build your own way to improve and scale work processes dramatically. And affordable.
The exercise at the heart of it
This is the “reshape” move that turns the vision into action, and it is the infinite-resources question in reverse: instead of asking what you would build with everything, you strip a function back to a blank sheet and ask what you would rebuild. In every established organisation, the instinct for getting better is to add capacity to the work you already do: hire more people, and train the people you have to do that same work better. Here, the first move is a thought exercise that runs the other way, and it can feel counterintuitive.
The exercise we recommend is not “who can we remove.” It’s: if we didn’t have these people, and we’re keeping them, we value them, how much of what they do could become agentic? The answer is usually “the vast majority.” But that’s only the start of the analysis. What you already know about the work those people do is the raw material for redesigning it. Once you’ve reimagined the work that way, you find the places where people are genuinely needed in this function, and that they’re needed for different work, at a higher level.
It’s an uncomfortable idea to put in front of middle management, the very layer the exercise reimagines, and we would not recommend opening with it. The better approach is to hold the radical end-state in mind while meeting an organisation where it actually is. Apply where it’s easy and obvious, Reshape the workflows deliberately, manager by manager, and signal to your fastest AI-embracers to Innovate early rather than holding it until the end. They are the ones who reveal which areas, and which people, the bigger agentic bets should come from.
Where we come in
We believe the organisations that thrive in a market increasingly full of agentic AI will be the ones whose leaders learned to start from the vision and then build step by step.
That’s the work we care about at Copenhagen Product Collective: helping leadership teams see the world the way agentic thinkers do, picture the version of their business worth building, and then take the realistic, sequenced steps to get there. If you’d like to run the infinite-resources exercise on your own organisation, and turn the answer into a plan, we’d love to talk.
At Copenhagen Product Collective, we work with companies to assess their product and technology leadership structures and build capabilities across both domains. If you’re wrestling with questions about how to structure your senior leadership and team topology, let’s talk.
