There is a fear running through the current discourse about AI that has the shape of something true but points in the wrong direction. The fear is that human intelligence is about to become useless — that when machines can think, the thinking humans do will stop mattering. I want to argue that this gets the causality exactly backwards. And I want to do it by looking at a moment we've already lived through.
- How word processors changed writing by making iteration cheap — and what skill that replaced
- The repeating pattern: every tool revolution removes an execution tax and moves craft upstream
- What happens to product management when the cost of building approaches zero
- The artist whose vision always outran technical capacity — and what the gap closing means for them
- What cognitive functions atrophy when tools externalize them, and which losses actually matter
- Why taste and judgment become more differentiating, not less, when execution becomes cheap
The Word Processor as Cognitive Prosthetic
Before word processors, writing was costly to revise. Not emotionally costly — physically, structurally costly. You committed words to paper and changing them meant cutting, pasting, retyping. The technology imposed a tax on iteration. Writers adapted by moving more of the composition process into their heads before their hands moved. You had to hold the argument, or at least a larger piece of it, in working memory before externalizing it. Some writers found this clarifying. Most found it constraining.
What the word processor changed wasn't writing. It changed the relationship between thinking and committing. Drafting became cheap, which meant you could think by writing rather than write after thinking. You could discover what you believed by putting something on the page, reacting to it, and revising. The iterative loop tightened dramatically.
The skill that was lost was composing in working memory — holding large structures mentally before committing them. The skill that was gained was something harder to name: the capacity to think in drafts, to treat the page as a thinking medium rather than a record of completed thought.
The tradeoff seems clearly worth it. The question was never whether writers could hold more in their heads. It was whether the writing got better. It did.
What Constraints Actually Constrain
Every tool revolution in creative and intellectual work follows the same underlying pattern. The constraint being lifted is always a tax on execution — some cost imposed between the intention and the artifact. Desktop publishing taxed layout. Digital audio production taxed recording and mixing. Cheap cameras taxed film. In each case, removing the tax didn't eliminate craft. It moved craft upstream.
When desktop publishing arrived, the fear was that it would flood the world with ugly documents. It did — temporarily. But it also enabled people with genuine aesthetic vision to execute that vision without a professional typesetter as intermediary. The signal-to-noise ratio got worse before it got better. The ceiling moved dramatically higher.
The pattern held for music. For film. For writing itself, when self-publishing removed the editorial gatekeeping function. In each case: a burst of low-quality output from people who mistook tool access for taste, followed by a period where people with actual vision discovered they could now execute it.
We are at the beginning of that same curve for software and for art. The AI slop critique is real — and it's a phase critique. It describes what happens when people use AI as a replacement for taste rather than an amplifier of it.
What Happens When the Binding Constraint Changes
Product management, as a discipline, is largely a formalized response to a specific scarcity: engineering time. The entire intellectual apparatus — roadmaps, prioritization frameworks, feature scoring, MVP definitions, the eternal negotiation between product and engineering — exists because building things is expensive. You have finite capacity, more possible features than you can build, and someone has to make the calls.
That constraint shaped not just the process but the thinking. PMs became skilled at arguing for features under resource constraints, at making tradeoffs legible to stakeholders, at deciding what to defer. A significant fraction of product thinking is actually constraint management dressed up as strategy.
When the cost of building a feature approaches zero, the intellectual problem changes entirely. It stops being: what do we build given what we can afford? It becomes: what is actually worth having? That is a harder question. And a more interesting one.
The shift pushes product thinking back toward genuine user understanding — toward the question of what creates real value — rather than toward the question of what clears the engineering bar. The PM who spent twenty years optimizing under scarcity may find that the scarcity was load-bearing for their entire mental model. The PM who always cared most about the user's actual problem is about to have a remarkable few years.
On Imagination Outrunning Technical Capacity
There is a category of artist, in every medium, whose imagination consistently outran their technical capacity. The filmmaker who could see the scene but not light it. The composer who could hear the orchestration but could not notate or produce it. The software designer who could envision the interface but not implement it. These people existed throughout history, and most of them experienced their limitation as a permanent condition. Talent without craft is frustrating in a specific, grinding way.
Technical constraint has also, of course, generated creativity. The three-chord song. The lo-fi aesthetic that became intentional. The 140-character tweet that became a literary form. Limitation as productive friction is real. Some artists built their entire practice around the resistance that constraint provided.
But these are two different populations. The artists who used constraint as generative material will adapt — they'll impose their own constraints, or find new ones. The artists whose vision was always larger than their execution capacity are about to discover something unprecedented: the gap closing.
Removing the execution tax doesn't homogenize creative output. It reveals what the output was always limited by. For some people, the limit was craft. For others, it was always vision. The ones whose limit was vision are about to find out what they were actually capable of.
The coming years are likely to see a burst of low-quality AI-assisted art from people who confused the tool with the talent. And alongside it, a body of work from people who always had the vision and now have the means — original, technically accomplished, and unmistakably theirs.
The Working Memory Problem
Every tool that externalizes a cognitive function changes the distribution of cognitive skills in the population that uses it. GPS weakened spatial navigation in people who relied on it heavily. Calculators reduced comfort with mental arithmetic. Word processors weakened, for some writers, the capacity to compose in working memory.
The honest account of AI-assisted software development has to acknowledge: some things will atrophy. The developer who always worked with AI assistance may develop weaker implementation intuition — the felt sense of what's hard to build, what breaks under scale, where the edge cases hide. That intuition accumulates from building things the hard way, from debugging problems you introduced yourself, from feeling the resistance of the material.
Whether that matters depends on a question worth taking seriously: was the intuition the point, or was it a byproduct of doing the actual work? If the work is getting the software to exist and behave correctly, and AI assistance reliably produces software that exists and behaves correctly, the intuition was instrumental. If the intuition is itself what makes good architectural decisions possible — if knowing what's hard is what allows you to design things that aren't — then losing it has downstream costs that won't be visible for years.
This is the honest tension. Not that AI makes us dumb. But that any tool which externalizes a cognitive function changes which cognitive functions we exercise. The question is always: which ones are worth preserving, and which ones were only necessary because we didn't have the tool?
What Actually Becomes More Valuable
The word processor didn't make writing easier in the sense of requiring less skill. It moved the skill. Composition in working memory mattered less. Judgment about what to cut, what to revise, what the argument actually was — that mattered more, because you could now afford to discover it through drafting rather than having to know it in advance.
Every tax-on-execution reduction works the same way. The skill doesn't disappear. It relocates upstream.
What relocates upstream when AI handles execution? Taste and judgment — the capacity to recognize quality — in software, in art, in argument — before it exists, well enough to direct something toward it. The ability to know what you want, which sounds trivial and is extraordinarily rare.
These capacities have always been valuable. What changes is the leverage they carry. When execution is expensive, taste is bottlenecked by resources. A person with impeccable taste and no technical skills can influence a small perimeter — the things they can personally execute, or the things they can afford to have executed. When execution becomes cheap, that perimeter expands dramatically. Taste becomes the binding constraint.
The person with great taste and no technical skills used to be frustrated. Now they're dangerous.
This is not an age of human obsolescence. It's an age in which the specifically human capacities — knowing what matters, recognizing quality, having genuine standards — become more differentiating, not less. The machine handles the gap between vision and artifact. The vision still has to come from somewhere.
That seems like the better problem to have.