The Capability Filter
April 21, 2026
I have terrible handwriting. Not "could improve with practice" bad -- genuinely, constitutionally bad. In elementary school it got me pulled aside and placed into a 'special' class. In university it hampered my work--essays on exams I had to write twice--once to get it down and a second time to write it legibly. Granted, I did do edits in between, but the reality is while others handed in 5-6 pages of handwritten essays I regularly asked for and got 2nd and 3rd notebooks. I was born lucky: typewriters were invented--I did a full year in Grade 9 Typing on a mechanical typewriter. By Grade 10 we were on WordPerfect and then WYSIWYG (What You See Is What You Get--visual word processors like we have now) before high school was done.
If I had been born a hundred years earlier, that would have mattered in ways it doesn't now. Clerical work -- the knowledge economy of the early twentieth century -- required legible script, correspondence, ledgers, reports. The pen was the interface and if you couldn't use it well, you were out regardless of what you actually had to contribute (unless, famously, you were a doctor). Not because the work required beautiful handwriting, but because the technology of the times did.
That's a filter: not a test of cognitive capacity, not a test of task completion, but a sheer capability filter.
It seems that lots of dominant technologies have imposed one.
Homer's world ran on memory. To carry knowledge -- to transmit stories, history, law, poetry -- you had to hold thousands of lines in your head. The Iliad is roughly fifteen thousand lines. If you couldn't memorize at that scale, you couldn't participate in the transmission of culture. Not because cultural insight requires perfect recall, but because the technology -- oral transmission -- demanded it. The filter selected for long form recitation, the task required judgement.
China's imperial civil service ran a version of the same filter for over a thousand years. The keju exam was explicitly meritocratic: any man could sit for it, regardless of birth. In practice, preparation required years of study that only wealthy families could fund, and success required calligraphy that took a childhood to develop and memorization of a specific Confucian canon. A man who governed his village effectively -- who understood people and power and consequence -- could fail the exam because his brushwork was wrong. The filter selected for exam-passing, the task required governance.
The print era imposed its own version. Literacy, yes -- but also a specific kind of literacy: linear, sequential, text-based. People who think in images, who jump between ideas, who need to talk through an argument rather than write it down, were at a systematic disadvantage in a world where written text was the primary interface for knowledge work. Not because their thinking was worse but because the interface didn't accommodate it. The filter selected for linear thinking, the tasks required invention.
The through-line: the filter selected for the interface skill, not the underlying capacity the task required.
How many capable people were filtered at these and other times in history merely because of the timing of their birth?
What AI actually did is specific.
It isn't that the tools got better or easier -- it's that the interface became fungible. TTS (text to speech), STT (speech to text), text-to-image, image-to-text -- multiple paths into the same creative work. If you can't draw, you can describe. If you can't type, you can speak. If you can't read musical notation, you can hum. If reading is hard, you can listen; if writing is hard, you can dictate. The interface requirement that used to be fixed is now a choice.
That's structurally different from previous tool improvements. A better brush is still a brush. A more intuitive drawing app still requires that you draw. A faster word processor still requires that you write. Each generation of tooling made the existing interface easier to use, but it didn't change which interface you were required to use. Multimodal AI doesn't lower the barrier -- it makes the barrier itself optional.
Consider the person who has carried a visual idea in their head for forty years -- a specific dragon soaring over a specific landscape -- and couldn't draw well enough to make it real. Before, the path from imagination to artifact ran through draftsmanship, and they didn't have it. Now the path runs through description. They tell a model what they see. The model produces it. They iterate. The dragon soars.
That's not the dragon being made for them. The dragon was always theirs. What changed is that the path between their head and a thing they can show another person doesn't go through a skill they don't have.
Creative vision has always been the thing, but the execution was the toll the old interface charged. AI stops charging that toll.
Not all barriers fall: cost, connectivity, the ability to articulate what you want, the cognitive work of knowing what's worth making -- these remain. What falls is a specific class: the interface-requirement filter. The one that said you must be able to do X to use this tool, where X had nothing to do with whether you had anything worth making at all.
The person with the thing in their head -- the image they couldn't draw, the story they couldn't get out because the prose came out wrong, the music they heard internally and couldn't transcribe -- that person was always there. The filter just meant we never saw what they were carrying.
The argument that a filter dropping is the same as something being stolen doesn't hold. Art isn't being taken away from artists, but the tax on entry is being removed.
What gets unlocked when a filter drops isn't guaranteed to be good. More creation means more noise. But it also means the ideas that lived and died inside people who couldn't execute them start moving.
The filter has dropped before. Print removed the memory requirement. Photography removed the draftsmanship requirement for visual documentation. Word processing removed the penmanship requirement for written thought. People still memorize their favourite stories. People still draw in their sketchbooks. People still do calligraphy for fun. The filter dropping takes nothing away from the artistry of those activities, the craft of their art is still there, but the technical requirement of that art is no longer a barrier to entry.
Each time, the craft survived, what didn't survive was the filter.
So, to everyone out there with their version of bad handwriting, pick up this new tool and see what the filters have been stopping you from creating.
James Henry is a Senior Security Consultant and independent AI/ML interpretability researcher. This post is about the filter no one talks about -- the one that selects on interface, not capacity. He still has terrible handwriting; the keyboard saved him. He writes about human-AI collaboration, digital sovereignty, and the tools that change who gets to build. He publishes at waypoint.henrynet.ca.