The three horizons: improve it, redesign it, or start again
When AI enters a business or a workflow, there are three distinct responses available. Choosing the wrong one does not just waste money — it means building the wrong thing at the wrong time.
When AI arrives in your work — or you arrive at AI — you face the same question every time: how much of what already exists should you keep? The answer is not a matter of enthusiasm or budget. It depends on where you are in the cycle and what the competitive landscape around you is doing.
There is a framework for this. It is not mine. I am going to be honest about where it comes from and then explain why it applies to AI decisions in a way its authors never anticipated.
The framework and where it comes from
The Three Horizons model comes from a 1999 book called The Alchemy of Growth, written by consultants at McKinsey. I encountered it working at PA Consulting and have used it ever since. The original idea was about strategic portfolio management — how companies should allocate attention and investment across their existing business, their growth initiatives, and their long-term bets simultaneously.
Applied to AI, it becomes something more immediate: a way of choosing how much of your existing work to change, and in what direction. The three horizons are not stages you move through in sequence. They are options available to you right now, and the choice between them has real consequences.
Horizon one: improve what already exists
Horizon 1 is the factory floor. The building is standing, the machinery is running, and you are looking at what can be made faster, cheaper, or better without structural change.
In AI terms: this is upskilling your staff, buying a Copilot licence, running workshops on better prompting. You are adding power tools to an existing process. The output improves, the speed increases, and the overhead stays roughly the same.
This is where most organisations are today. It is also the right starting point — you cannot jump to Horizon 3 from nothing, and the knowledge you gain from improving a familiar process is what makes the next horizon navigable. Horizon 1 earns its place.
The limitation is the curve. Horizon 1 investments deliver value quickly, look good in early reviews, and then slow. The improvement is bounded by the original process. You are running faster on the same track, and the track has an end.
The point at which Horizon 1 stops being enough is when competitors who have invested in Horizon 2 or 3 start to move faster than the incremental gains you are making. That inflection is not always visible until you are already behind it.
Horizon two: redesign the process
Horizon 2 is not about making the floor faster. It is about looking at the floor and asking whether the floor should be configured differently.
In the factory analogy: you are bringing in a robotic arm, a new conveyor system, specialists trained in process optimisation. The workers are still there, but their jobs have shifted — they are managing the machinery rather than operating it. The process has been redesigned from within.
This is what a management consulting firm is typically hired to do: map the current state, model a future state, identify the gap, and design a path between them. Gap analysis is Horizon 2 work. It is more expensive than Horizon 1, takes longer to show results, and holds its value for longer — because the new process was built for the current landscape rather than adapted from a previous one.
In AI terms: Horizon 2 is redesigning a workflow around AI's actual capabilities rather than bolting AI onto the existing steps. You are asking not "how does AI speed this up?" but "if we were designing this workflow today, knowing what AI can do, what would it look like?" Those are different questions. The second one often produces a process that looks nothing like the first.
It's very much the work of a management consulting firm to come in and look at the process and suggest a new one. You look at the current state, map it out, how that looks, and then suggest a future state, and then you show them the gap.
Chapter 1.5 · 03:42
Horizon three: build from scratch
Horizon 3 is first principles. The question is no longer how to improve or redesign — it is what you would build if you were starting today, with the best available tools, unconstrained by what already exists.
Tesla's Gigafactories are the example I use in the chapter. They were not improved factories or redesigned factories. They were factories designed from scratch around the capabilities of modern robotics and manufacturing science, with every assumption revisited. The cost structure, the energy use, the quality standards, the floor layout — none of it was inherited from the previous model.
In AI terms: Horizon 3 is building a system where AI is not an addition to how work gets done, but the foundation of how it gets done. The AI-native version of the business. Processes that would have required a team of ten are designed for one person and a set of agents. Products that would have required physical delivery are delivered as software. Services that would have required a human specialist are automated end to end.
Horizon 3 investments take longer to produce value — the change management alone is significant — but once they are working, they compound. The value does not flatten in the way Horizon 1 does. And crucially, Horizon 3 systems often compete on a different axis entirely, which means the comparison to Horizon 1 or 2 competitors becomes meaningless.
The wrong horizon is its own kind of mistake
The mistake I see most often is treating Horizon 1 work as a strategy when the competitive environment has already moved to Horizon 2. You run faster workshops, buy more licences, prompt more carefully — and the output improves, which makes it look like progress, right up until a competitor who redesigned the process entirely takes the work you were doing in a week and does it in an afternoon.
The opposite mistake also exists: jumping to Horizon 3 before you understand your own processes well enough to design the replacement. First-principles thinking requires knowing what the principles actually are. If you have not done Horizon 1 or 2 work, you do not have that knowledge yet.
Horizon 3 is not always the answer. Some businesses genuinely benefit from a well-executed Horizon 1 improvement and do not need to go further. The decision depends on how competitive the environment is, how fast it is moving, and how much the current process is actually the constraint.
How the horizons interact
The horizons are not mutually exclusive. Most healthy businesses are running all three simultaneously — defending and optimising the current operation, redesigning specific workflows, and making early bets on what the next version of the business looks like.
The course structure reflects this. The act called Everyday Magic covers Horizon 1 work: taking familiar tasks and doing them better with AI. The act called Sorcery covers Horizon 2: redesigning workflows. The act called God Mode covers Horizon 3: building new systems from the ground up. The horizons are a map of the course because they are a map of the decision.
What usually goes wrong
The most common failure mode is using Horizon 1 language to describe Horizon 3 ambitions. "We're going to use AI to transform how we work" is a Horizon 3 ambition. "We're running AI workshops for staff" is Horizon 1 execution. The gap between those two things is not a communication problem — it is a planning problem, and no amount of better workshops closes it.
The second failure mode is treating the horizons as a linear progression you must complete in order. You do not have to wait until Horizon 1 is finished before starting Horizon 2. In a fast-moving environment, waiting is itself a decision with consequences.
Choosing
The useful question is not "which horizon is best?" It is: given where my business is, where my competitors are, and what the next 18 months of this market look like, which horizon is the right frame for this specific decision?
That is a judgment call. The model does not make it for you. What it does is stop you from defaulting to the horizon that feels most comfortable, rather than the one that is actually right.
Drawn from chapter 1.5 of AI Magic 2033.
