How to Expand Cognitive Work and Build AI Literacy

This 90% gap tells us something crucial about where we are on the AI technology adoption curve and it’s not just about the numbers.

What strikes me most isn’t the 10% penetration figure, but what those 10% are actually doing with ChatGPT.

The dominance of writing and practical guidance (56% combined) reveals something we constantly see in organizational adoption: people aren’t looking for magic. They’re looking for partnership in tasks they already understand but want to do better or faster.

The relatively small share of coding (8%) is particularly interesting for those of us working on enterprise AI adoption. The narrative around AI has been dominated by technical use cases, yet consumer behavior shows that language work—editing, communication, translation, tutoring—represents where real cognitive effort happens for most people.

This gap between corporate expectations and actual usage patterns is something organizations must confront as they build AI strategies.

But here’s the deeper meaning: the remaining 90% won’t simply replicate these patterns at scale. As adoption moves from early adopters to populations with different educational approaches, literacy levels, and technological comfort, we’ll likely see an explosion in the “Practical Guidance” category.

Tutoring, how-to advice, health guidance this represents the democratization of access to expertise that has historically been limited by cost, geography, or credentials.

The minimal use for self-expression (4%) is encouraging from a capability-building perspective. It suggests most users instinctively understand the difference between a tool and a relationship though this may change as adoption reaches isolated or emotionally under-supported populations.

For those of us designing AI adoption frameworks, this data confirms a crucial point: sustainable AI integration isn’t about teaching people to do fundamentally new things.

It’s about expanding the cognitive work they already do writing, learning, problem-solving and building the literacy to do it well. The 90% opportunity isn’t just market growth.

It’s a responsibility to ensure that expansion happens with capability transfer, not dependency creation.

How to Expand Cognitive Work and Build AI Literacy

The key is contextual scaffolding—gradually building skills through authentic work tasks, not through abstract “how to use AI” training.

Three-Layer Approach:

1. Task as Entry Point (Not the Tool)

• Don’t start with “here’s ChatGPT, learn to use it”
• Start with: “you have a report to write,” “you have an email you don’t know how to formulate,” “you’re trying to understand a complicated procedure“.
• AI becomes a tool within familiar context, not a new skill to master.

2. Explicit Learning of Thinking Structure

• People don’t know what they don’t know about their own cognitive process.
• When using AI for writing, they must understand: how to structure an argument, how to check logic, how to recognize the gap between “sounds good” and “actually makes sense”.
• This means learning metacognition through LLM interaction. AI becomes a mirror of their thinking.

3. Iterative Independence

• Phase 1: AI generates, human critiques (develops quality criteria)
• Phase 2: Human gives AI instructions, AI executes (develops precision in communication)
• Phase 3: Human and AI co-create (develops strategy and judgment)
• Phase 4: Human works independently, uses AI for verification (internalized skills)

Democratization of Expertise: How the Transition Happens

Here’s how this transition unfolds through concrete mechanisms:

1. From Explicit to Tacit Knowledge

Historically, expertise has been accessible in two forms:

• Explicit knowledge: books, courses, certificates (cost-prohibitive).
• Tacit knowledge: experience, mentorship, “office hours with an expert” (geographically and socially limited).

AI assistants change this equation because they can:

• Provide personalized explanation of complex concepts (like a private tutor).
• Answer “stupid” questions without judgment (social barrier falls).
• Give context-based advice without needing expensive consultants.

Example:

• Before: To learn how to conduct difficult conversations with employees, a manager needed either expensive training or a good mentor.
• Now: Manager can simulate the conversation with AI, get feedback on approach, iterate until gaining confidence.

2. Transition to Less Technically Literate Users

Early adopters (10%) have two advantages:

• Technological literacy (know how to formulate prompts, how to iterate).
• Critical thinking (know when AI is wrong, have expertise to verify).

The next 90% often lack both. That’s why intermediary structures must exist:

a) Predefined Scenarios Instead of Empty Interface

• Not: “Here’s ChatGPT, use it however you want”.
• Yes: “Click here for: help writing your CV / understanding a medical report / budget planning”.
• This is why custom AI agents and specialized interfaces become critical.

b) Built-in Safeguards Against Misuse

• AI that says “for this you should consult a real doctor” instead of giving medical diagnoses.
• AI that explicitly shows uncertainty instead of false authority.
• Different access levels depending on user knowledge.

c) Social Infrastructure Around the Tool

• AI adoption must have a learning community.
• Peer-to-peer sharing of use cases (“here’s how I used it”).
• Local “AI literacy workshops” that translate technology into the language of people’s real problems.

3. Specific Categories That Will Explode

Tutoring (10% → potentially 25%+)

• Personalized learning of anything—languages, skills, academic subjects.
• Available 24/7, no cost, adapted to learning pace.
• Eliminates geographic and economic barriers to education.
• Critical component: teach people TO ASK THE RIGHT QUESTIONS, not just accept answers.

How-to Advice (9% → potentially 20%+)

• From fixing a faucet to starting a business
• Replacement for expensive consultants, YouTube searches, and “someone who knows someone who knows”.
• Risk: overconfidence in areas where accuracy is critical (construction, health, law).
• Verification system needed—AI + human expert for critical areas.

Health and Self-Care (6% → potentially 15%+)

• Interpreting symptoms (not diagnoses), understanding treatment options, navigating healthcare systems.
• This is where democratization is most important (health inequalities are enormous) but also riskiest.
• Model: AI as information layer BEFORE professional consultation, not replacement for it.

4. What Must Happen for This to Be Sustainable

Media and AI Literacy as a Fundamental Skill

• Just as writing and arithmetic were fundamental skills of the 20th century, the ability to work with AI is a fundamental skill of the 21st century.
• This means: understanding what AI can/cannot do, how to verify output, how to iterate.
• It can’t be “learn ChatGPT”—it must be “learn how to think WITH AI”.

Cultural Adaptability

• AI systems trained on Western data don’t understand the context of other cultures.
• Need capacity for AI to “adapt” to local norms, language, values.
• This is why custom AI agents and local implementations are superior to monopolistic platforms.

The next 90% of users won’t be data scientists. They’ll be nurses needing help with documentation, teachers wanting customized lesson plans, local entrepreneurs trying to understand their market.

Our job is to ensure that when that wave comes, people gain capabilities, not just access.