AI Is Not a Tool. It Is a Catalyst.
September 12, 2026
AI Is Not the Dynamite
What happens when intelligence becomes abundant?
I recently saw an image of AI as a person with an axe, cutting into the foundations of the software industry.
It is a good image.
But I think the axe is the wrong metaphor.
An axe makes the miner faster.
Dynamite changes the mine.
That distinction is important.
The mine
For decades, software has been a kind of mine.
We have ideas.
We have visions.
We have problems we want to solve.
But between the idea and the thing that actually works is a large amount of human effort.
Someone has to understand the problem.
Someone has to design the system.
Someone has to write the code.
Someone has to test it.
Someone has to fix it when reality disagrees with the design.
The developer is the miner.
Code is what gets extracted.
And human time is the resource being consumed.
AI changes the economics of that extraction.
Not because it writes code faster.
That is the least interesting part.
The interesting part is that the amount of human time required to turn an intention into a working system may be falling.
That is a much bigger change.
Catalysts
I have started to think about technological progress differently.
Not as a line.
More like a network.
One technology enables another.
The steam engine did not simply become a better machine.
It changed what could be built.
Electricity changed what machines could do.
Computers changed the economics of calculation.
The Internet changed the economics of communication.
Dynamite changed the economics of extraction.
These technologies have something in common.
They are catalysts.
A catalyst does not necessarily create something that was impossible before.
It changes the conditions.
It lowers the barrier.
It accelerates the reaction.
And once something becomes sufficiently cheap or fast, people start doing much more of it.
That last part matters.
Because catalysts do not merely make the same world more efficient.
They can create a different world.
AI as a catalyst
AI may be doing something similar to cognitive work.
Writing becomes cheaper.
Coding becomes cheaper.
Analysis becomes cheaper.
Research becomes cheaper.
Design becomes cheaper.
The scarce resource underneath all of these activities is largely the same:
time.
If something that took three weeks takes three hours, the important question isn’t whether the AI is a good programmer.
The question is what happens when three weeks of cognitive work no longer costs three weeks of human time.
And then what happens when that makes the next thing economically viable.
And the next.
And the next.
That is where AI becomes interesting.
Not as the endpoint.
As an accelerator of the process that produces what comes next.
The strange thing about software
There is a problem, though.
AI is very good at producing the first version of something.
The difficult part may be everything that comes afterwards.
A product contains years of decisions.
Dependencies.
Assumptions.
Compatibility.
Security constraints.
Technical debt.
Knowledge that nobody ever wrote down.
The older the product, the more of that history is embedded in it.
AI has to excavate the mine before it can change what is inside it.
Which leads to a thought I keep coming back to:
AI is expensive when the input is a product, not a vision.
A product contains implementation history.
A vision contains intent.
That difference may become increasingly important.
Perhaps the future of software is not maintaining implementations indefinitely.
Perhaps it is maintaining the vision, constraints, invariants and tests — and allowing AI to continuously regenerate the implementation.
Not:
Build → maintain → patch → refactor → maintain.
But:
Vision → build → observe → correct → rebuild.
The code becomes temporary.
The intent becomes permanent.
The dynamite problem
But here the metaphor becomes uncomfortable.
Dynamite made extraction cheaper.
It did not necessarily make humanity extract less.
It made more extraction economically attractive.
And more extraction has consequences.
Water.
Landscape.
Waste.
Energy.
Ecosystems.
The same could happen with AI.
If intelligence becomes cheap, we may not use less intelligence.
We may use vastly more of it.
More software.
More simulations.
More agents.
More experiments.
More computation.
More data centers.
More electricity.
More heat.
Efficiency does not automatically mean less consumption.
Sometimes efficiency makes consumption explode.
The cheaper something becomes, the more of it we do.
So perhaps AI is not the dynamite.
Perhaps AI is the catalyst for the dynamite.
What comes next?
This is the part I find most interesting.
Imagine AI becomes capable enough to accelerate scientific discovery itself.
It designs experiments.
Runs simulations.
Develops materials.
Designs machines.
Finds engineering solutions.
Builds better AI systems.
Those systems accelerate the process again.
AI → better tools → better science → better materials → better machines → better AI.
A loop.
At some point, something we cannot currently name may emerge from that loop.
A new material.
A new manufacturing principle.
A new energy technology.
A new kind of machine.
Something that changes the economics of the physical world in the same way that software is beginning to change the economics of cognitive work.
We don’t know what it is.
And I think pretending that we do would miss the point.
The interesting question is not:
What will AI become?
It is:
What will AI make possible that we currently cannot afford to discover?
From idea to object
Now take that thought to an extreme.
Imagine a future in which AI helps humanity discover a universal material.
And a machine capable of turning that material into almost anything.
Neither exists today.
This is not a prediction.
It is a thought experiment.
But consider what happens if it becomes possible.
Today:
Idea → research → design → engineering → materials → suppliers → factory → logistics → assembly → product
Imagine that chain collapsing.
Toward:
Vision → object
Suddenly manufacturing is no longer the dominant constraint.
And that changes the question.
Today we ask:
Can we build it?
Tomorrow we might ask:
Can we afford to build it?
In an omni-production world, perhaps we eventually ask something else:
Should we build it?
When cost stops being interesting
This is where I think the word cost starts to lose some of its explanatory power.
Cost is useful when production is scarce.
But imagine the marginal cost of producing many things approaching zero.
The interesting question is no longer:
How cheaply can we make this?
It becomes:
What do we get in return?
Return on investment.
Return on resources.
Return on energy.
Return on attention.
Perhaps even:
Return on intention.
Did we move the world closer to something we actually wanted?
Because there is a strange consequence to abundance.
If almost anything can be produced, production is no longer the main bottleneck.
Selection is.
The new scarcity
Today, much of our world is constrained by what we can produce.
Not enough engineers.
Not enough factories.
Not enough materials.
Not enough energy.
Not enough time.
Technological progress repeatedly attacks these constraints.
Dynamite reduced the human effort required for extraction.
Computers reduced the cost of computation.
The Internet reduced the cost of moving information.
AI is reducing the cost of cognitive production.
A future manufacturing technology could reduce the cost of physical production.
And then what?
Perhaps the scarce resource becomes something much harder to manufacture:
a reason.
A reason for something to exist.
The abundance problem
A world where everything is cheap is not automatically a better world.
If something costs almost nothing to produce, there is little stopping us from producing far too much of it.
More objects.
More software.
More computation.
More noise.
More complexity.
More things that nobody actually needed.
The catalyst does not know whether the reaction is good.
It only makes the reaction easier.
That is the uncomfortable part.
Technology gives us leverage.
It does not give us purpose.
The shift
Maybe this is the deeper pattern behind technological progress.
First, we ask:
Can we do it?
Then:
Can we do it efficiently?
Then:
Can we do it cheaply enough?
And eventually, perhaps:
Should we do it at all?
The bottleneck moves.
From production.
To efficiency.
To abundance.
And finally to choice.
That is a very different world.
AI is not the destination
This is why I am increasingly reluctant to think of AI as the final technological revolution.
Maybe it is not.
Maybe AI is another catalyst.
A catalyst that accelerates the discovery of things that will make AI itself look like a transition technology.
We may be watching the beginning of a chain rather than the arrival of its destination.
And that makes the most interesting question surprisingly simple:
What does AI make possible next?
I don’t know.
Nobody does.
But perhaps that is exactly why it is worth thinking about.
Because the future may not belong to the technology that becomes the most powerful.
It may belong to the technology that makes the next catalyst possible.
And if we eventually reach a world where turning a vision into reality becomes almost effortless, the hardest problem may no longer be intelligence.
It may be deciding what is worth wanting.