Analogue
before LLM.
Develop human capability first.
Then amplify it.
We have a tendency to begin with the finished thing.
The answer.
The skill.
The innovation.
The technology.
We evaluate what something can do without always asking what had to happen for that capability to exist. That distinction matters enormously in the age of artificial intelligence.
Development
before output.
AI can produce remarkable outputs. It can accelerate research, organize complexity, surface patterns, generate possibilities, and take on work that consumes human time without necessarily requiring human judgment. Those capabilities can create tremendous value, but if we begin with the output, we may miss the more consequential question:
The finished thing
can conceal the system.
In September 2026, Stanford Medicine researchers reported something remarkable about an organ we might assume we already understood: the human brain.
Their work found that the forebrain and midbrain arise from a different progenitor-cell lineage than the hindbrain. Rather than beginning as one developmental system that later divides, the two follow distinct developmental paths from remarkably early stages.
One of the insights behind the discovery was also striking. Stem-cell research had often focused on producing the desired end cell type. This team looked further upstream toward the earliest stages of development.
This research is not evidence about AI, education, or analogue learning, but as a systems thinker, I find the parallel difficult to ignore.
Sometimes the finished thing tells us far less about how capability develops than we imagine.
Don't look only at the output. Understand what had to develop to make the output possible.
Start with
the lump of clay.
Perhaps that is one reason I keep returning to analogue experiences—not because analogue is inherently superior to digital, but because some of our earliest encounters with the world ask something important of us before giving us an answer.
Give a child a lump of clay or Silly Putty and there is no interface explaining what to do next. No tutorial demonstrating what the finished thing should look like. No algorithm offering another example before the child has formed an idea of their own.
So they squeeze it. Stretch it. Pull it apart. Put it back together. Make something ridiculous. Discover that one idea doesn't work. Try another.
What looks like play is also practice.
These are creative and critical-thinking muscles. Children need opportunities to exercise them before technology becomes responsible for supplying the possibilities.
A video can show a child what someone else discovered and what could be made from a lump of clay; that can be valuable, but it is not the same experience as discovering possibility before someone else defines it.
Teaching Others
to See.
I've seen a version of this much later in life, too.
In my Threshold's "Teaching Others to See," Ariel began to participate, explore, and see possibilities in herself that had not been visible before. The important thing wasn't simply what she ultimately produced. It was what began developing through the process.
We shouldn't confuse exposure to someone else's thinking with the development of our own.
The hand
is thinking, too.
Research comparing handwriting and typewriting provides another useful window into the distinction between completing a task and understanding what the task asks of us.
In a high-density EEG study of 36 university students, researchers recorded brain activity using a 256-sensor array while participants wrote words by hand and typed them on a keyboard.
Handwriting produced substantially more elaborate connectivity patterns across brain regions associated in existing research with memory formation and the encoding of new information.
The point is not that keyboards should disappear. Nor does one study settle how every person learns best.
It reminds us that two activities producing the same visible result—a written word—may ask very different things of the person producing it.
Source: Van der Weel & Van der Meer, Frontiers in Psychology
Not all friction
is a problem.
Some friction consumes human capacity. Some friction develops it. Knowing the difference may be one of the most important design decisions we make. The goal isn't to preserve inefficiency. It is to understand which effort is doing developmental work before we automate it away.
Unnecessary friction
Repetitive administration
Retrieval
Formatting
Routine comparison
Organization
Low-risk repetitive analysis
Developmental friction
Questioning
Reasoning
Making
Struggling
Evaluating
Imagining
Faster isn't the
same as learned.
That distinction is becoming increasingly important as generative AI enters education.
The OECD's 2026 Digital Education Outlook reviews emerging evidence showing that general-purpose generative AI can improve performance on a task without necessarily producing corresponding learning gains.
Cognitive offloading can be useful. Humans have always used tools to extend memory, calculation, communication, and physical capability.
But when the cognitive activity being offloaded is itself part of what someone needs to learn, efficiency can become a poor proxy for development.
The OECD recommends something strikingly compatible with this idea: develop independent thinking and foundational skills, use educational AI intentionally, and use general-purpose AI selectively rather than allowing it to replace cognitive effort.
Source: OECD Digital Education Outlook 2026
THE DISTINCTION: Creating capacity for thought is very different from replacing thought.
Before a machine
could learn from it...
Large language models were built upon an extraordinary inheritance of human creativity, discovery, writing, art, code, problem-solving, experimentation, and innovation.
The technology is remarkable, so is the human inheritance that made it possible.
The question
is sequence.
The lesson isn't that the iPad is bad, the video is bad, or AI is bad. Each can open doors that a lump of clay cannot.
AI is a tool. When leveraged rationally and realistically, it can provide tremendous value: taking on repetitive tasks, accelerating research, enhancing ideas, and creating capacity for people to do more meaningful work, but human development should never become collateral damage of any tool.
We already create standards and safeguards when technologies have the potential to affect people and society at scale. AI deserves serious stewardship, particularly when it intersects with children, education, human agency, creativity, work, and the resources required to sustain increasingly powerful systems.
UNESCO's guidance for generative AI in education similarly calls for a human-centered, age-appropriate approach that protects human agency while developing people's capacity to use these systems meaningfully.
Source: UNESCO — Guidance for Generative AI in Education and Research
Analogue before LLM.
Perhaps that is
the better question.
The goal shouldn't be to extract everything AI is capable of giving us. Perhaps it is to determine where AI creates the greatest value for humanity while deliberately preserving—and continuing to develop—the distinctly human capabilities that made it possible in the first place.
Before we ask what technology can do for us, we might ask what we still need to learn to do for ourselves.
We shouldn't be teaching future generations to compete with AI. We should be teaching them to think deeply enough to know what to do with it.
