From End-User Computing to End-User AI: Why Pattern AI May Be the Missing Human Layer
Executive Summary
A major shift happened in computing several decades ago.
Computers moved from the exclusive domain of programmers, data-processing departments, and specialists into the hands of ordinary professionals. Managers built spreadsheets. Accountants created financial models. Engineers analysed data. Teachers prepared learning materials. Small businesses maintained databases.
The phenomenon came to be known as End-User Computing (EUC).
Today, Artificial Intelligence appears to be undergoing a remarkably similar transition.
AI was once primarily something that specialists built. Increasingly, it is something that millions of people use directly.
Students use AI to learn.
Teachers use it to design learning experiences.
Managers use it to analyse situations.
Professionals use it to draft, investigate, compare and decide.
Entrepreneurs use it to explore markets and develop ideas.
This suggests an important emerging distinction:
AI Engineering builds AI. End-User AI uses AI to accomplish something.
If End-User AI (EUAI) becomes a useful way of describing this new relationship between humans and intelligent systems, another question immediately follows:
What competencies does the end user need?
This is where Pattern AI can be positioned.
Pattern AI does not need to be presented as another AI technology or another AI model. It can instead be understood as an end-user approach to AI: a way of helping ordinary users understand problems, recognize patterns, communicate effectively with AI, examine the patterns in AI's responses, exercise judgment and convert AI assistance into real-world accomplishment.
The historical parallel is compelling:
Computing → End-User Computing
and now potentially:
Artificial Intelligence → End-User AI
with:
Pattern AI → a human thinking and interaction framework for End-User AI.
1. There Was a Time When Computing Belonged to Specialists
It is easy to forget how centralized computing once was.
During the mainframe era, computing resources were expensive, scarce and largely controlled by specialist departments. Business users typically did not sit down at a computer and directly construct solutions to their everyday problems.
They submitted requirements.
Programmers translated those requirements into programs.
Computer operators ran the programs.
Results eventually came back to the users.
The person who understood the business problem and the person who instructed the computer were often different people.
Then computing began moving toward the user.
Personal computers, interactive software, spreadsheets, database tools and other applications progressively changed who could make computers useful.
One particularly important milestone was the electronic spreadsheet.
VisiCalc, released for the Apple II in 1979, became one of the early landmark personal-computer applications. Instead of asking a programmer to create a program every time a financial assumption changed, a business user could construct a spreadsheet model and change the assumptions directly.
That sounds ordinary today.
At the time, it represented a profound change in the relationship between people and computers.
The computer was no longer merely something that specialists programmed for the user.
Increasingly, it became something the user could operate and shape for themselves.
2. The Rise of End-User Computing
By the early 1980s, researchers were already examining this phenomenon systematically.
John F. Rockart and Lauren S. Flannery's influential work on the management of end-user computing, published in Communications of the ACM in 1983 following earlier MIT research, documented the growing organizational importance of computing performed directly by users rather than exclusively through centralized information-systems specialists. (ACM Digital Library)
Over time, the broader field expanded into ideas such as End-User Programming and End-User Development.
The basic principle was powerful:
People should not necessarily have to become professional software developers before they can make computers solve problems relevant to their own work.
Research on End-User Development describes it as enabling non-professional developers to create, modify or extend computational artifacts. ACM literature similarly describes EUD as involving non-programmers tailoring or creating software artifacts, often within organizational settings. (ACM Digital Library)
Bonnie Nardi's influential book A Small Matter of Programming examined end-user programming through systems such as spreadsheets and computer-aided design, focusing specifically on the abilities and difficulties of users rather than treating programming solely as the territory of professional programmers. (MIT Press)
This was not simply a change in software.
It was a change in agency.
3. The Spreadsheet Was More Than a Tool
Consider what the spreadsheet actually accomplished.
Before spreadsheets, someone working in finance might formulate a problem and then depend on specialists or manual processes to calculate different scenarios.
With spreadsheets, the domain expert could manipulate the model directly.
The accountant did not have to become a computer scientist.
The computer adapted more closely to the accountant's way of working.
That distinction matters enormously for understanding today's AI transition.
The spreadsheet reduced the distance between:
Problem → Computer → Result
Generative AI reduces that distance even further.
Instead of requiring formulas, menus, programming syntax or carefully structured database queries, modern AI systems can increasingly be instructed through ordinary language.
We may therefore be witnessing something larger than the adoption of another software category.
We may be witnessing another major stage in end-user computing itself.
4. From End-User Computing to End-User AI
The historical progression can be simplified as:
Mainframe Computing
↓
Personal Computing
↓
End-User Computing
↓
Internet & Cloud Computing
↓
Mobile Computing
↓
End-User AI
The last term should be used carefully.
“End-User AI” is not yet a single universally standardized academic category equivalent to the mature literature surrounding End-User Computing.
However, the underlying phenomenon is very real.
Research already explicitly connects AI with End-User Development. A 2023 systematic literature review titled End-User Development for Artificial Intelligence examined approaches that allow people without AI or programming expertise to customize AI behaviour for their own needs. (arXiv)
More recent research is exploring AI-assisted end-user programming and Generative-AI-augmented End-User Development, suggesting that generative AI may substantially change how non-specialists create computational solutions. (arXiv)
So the terminology is still developing, but the direction is unmistakable:
AI capability is moving closer to the person who actually has the problem.
That is precisely what happened with End-User Computing.
5. What Exactly Is End-User AI?
I propose the following working definition:
End-User AI is the direct use, adaptation and collaboration with artificial intelligence by people whose primary role is not developing AI systems, in order to accomplish learning, professional, creative, analytical or everyday tasks.
The important phrase is:
whose primary role is not developing AI systems.
A teacher does not have to understand transformer architecture before using AI to develop examples for a lesson.
A marketing professional does not have to train a language model before using AI to compare campaign concepts.
A student does not need to implement backpropagation before asking AI to explain why K-Nearest Neighbours classifies a particular point in a certain way.
A manager does not need to become a data scientist before using AI to examine alternatives.
These people are not necessarily AI developers.
They are becoming AI end users.
6. AI Engineering and End-User AI Are Different Layers
This distinction helps organize the increasingly confusing AI landscape.
AI Engineering
Concerned with things such as:
Data → Models → Training → Evaluation → Deployment → Infrastructure
The central question is:
How do we build reliable AI systems?
End-User AI
Concerned with:
Problem → Interaction → AI Assistance → Evaluation → Application
The central question becomes:
How do I use AI intelligently to accomplish something?
Both matter.
But they require different competencies.
A doctor using AI does not necessarily need the same skills as the engineer developing the medical AI system.
A professor using AI does not necessarily need the same competencies as someone training a foundation model.
A business manager working with an AI assistant does not necessarily need to know how to implement attention mechanisms.
But all three need to know something extremely important:
how to think while working with AI.
7. Evidence That This Shift Is Already Happening
The movement toward End-User AI is not merely philosophical.
Workplace research provides evidence that generative AI is already being integrated directly into people's everyday tasks.
A major field study involving customer-support agents found that access to a generative-AI conversational assistant increased productivity by approximately 14% on average, with particularly substantial gains among less experienced and lower-skilled workers. (National Bureau of Economic Research)
Another experiment involving consultants found what researchers called a “jagged technological frontier.” For tasks within AI's capabilities, participants using AI completed more tasks, worked faster and produced higher-quality results. The study reported that AI users completed 12.2% more tasks and were approximately 25% faster. (Harvard Business School)
But there was a crucial warning.
For a task outside that frontier, AI users were more likely to arrive at an incorrect solution; Harvard's summary reports that participants using AI on the more difficult task were 19 percentage points less likely to produce the correct answer. (Harvard Business School Library)
This tells us something fundamental.
The important capability is not merely:
Can you use AI?
It is:
Can you determine when, where, why and how to use AI—and can you judge what it gives you?
That is an End-User AI problem.
8. The New Bottleneck Is Human Judgment
Traditional computing often demanded precision before execution.
You wrote the formula.
You specified the command.
You wrote the code.
If the syntax was wrong, the system often failed visibly.
Generative AI behaves differently.
A user can provide an incomplete question and still receive a fluent, polished response.
This creates a new problem.
Fluency can look like correctness.
Therefore the reduction in technical barriers does not eliminate the need for expertise.
It changes the kind of expertise required.
The end user increasingly needs:
problem-framing ability + domain knowledge + AI literacy + critical evaluation + judgment.
This is why AI literacy is receiving growing institutional attention.
UNESCO's AI Competency Framework for Students identifies 12 competencies across four dimensions, including a human-centred mindset, ethics, AI techniques and applications, and AI system design. It explicitly aims to prepare students to become responsible users and co-creators of AI. (UNESCO)
The OECD similarly describes AI literacy through competencies involving engaging with, creating with, managing and designing AI, rather than reducing AI literacy to knowing how to write prompts. (OECD)
The direction is clear:
Access to AI is not enough.
People need a way of thinking about AI.
9. This Is Where Pattern AI Enters
This creates an interesting role for the Pattern AI approach.
Pattern AI starts from a simple observation:
AI is fundamentally a pattern technology.
Machine-learning systems learn statistical regularities from data and use those learned relationships to classify, predict, retrieve, recommend or generate outputs.
For the end user, however, understanding every mathematical detail of those systems is neither always necessary nor sufficient.
The practical question becomes:
What pattern is operating here?
A student learning classification can look for:
Features → Similarities/Differences → Decision → Class
A manager analysing declining sales can investigate:
Change → Recurrence → Segment → Possible Cause → Evidence
A teacher designing learning can look for:
Concept → Misconception → Example → Feedback → Improvement
A recruiter can examine:
Role → Tasks → Skills → Evidence → Candidate Fit
A doctor might investigate:
Symptoms → Observations → Possible Patterns → Evidence → Professional Judgment
The domain changes.
The underlying reasoning often contains recognizable structures.
Pattern AI attempts to make those structures visible.
10. Pattern AI Is Not Another AI
This distinction is critical.
Pattern AI should not be presented as a new machine-learning architecture, foundation model or technical branch of Artificial Intelligence.
It is better positioned as:
An end-user approach for understanding, interacting with and applying AI through patterns.
Or more specifically:
Pattern AI is a human thinking framework for End-User AI.
This positioning avoids competing with technical AI.
Instead, it addresses a different layer.
Consider the stack:
AI Infrastructure
Compute, cloud, chips
↓
AI Models
Machine learning, foundation models, LLMs
↓
AI Applications
Assistants, copilots, agents, domain applications
↓
End-User AI
People using AI to accomplish real tasks
↓
Pattern AI
A framework helping the human understand, interact, evaluate and apply
The last layer is fundamentally about human agency.
11. The End User Needs More Than Prompt Engineering
The rapid rise of generative AI produced enormous interest in prompt engineering.
Prompting matters.
But reducing End-User AI competency to prompting would be like reducing computer literacy to knowing keyboard shortcuts.
A prompt is an interaction mechanism.
It is not the entire thinking process.
Before asking AI anything, the user must determine:
What am I actually trying to accomplish?
After AI responds, the user must determine:
Is this useful?
Is it correct?
What assumptions did it make?
What evidence supports it?
What is missing?
Should I act on it?
Therefore a mature End-User AI framework needs to surround prompting with human reasoning.
12. Think → Ask → Judge
This leads naturally to a simple operating discipline:
THINK
Before involving AI:
What am I trying to understand, decide, create or accomplish?
Identify the problem.
Look for the relevant pattern.
Bring context and domain knowledge.
↓
ASK
Collaborate with AI.
Provide context.
Ask for explanation, alternatives, comparisons, evidence or critique.
Iterate instead of accepting the first output.
↓
JUDGE
Evaluate what AI produces.
Check accuracy.
Challenge assumptions.
Verify important claims.
Apply domain knowledge.
Decide whether the result is fit for purpose.
This creates a much healthier relationship than:
ASK → COPY → USE
The desired relationship becomes:
THINK → ASK → JUDGE → ACCOMPLISH
That is a useful operating model for End-User AI.
13. Pattern AI Adds a Layer Inside This Cycle
Pattern thinking can operate throughout the process.
THINK
What pattern am I dealing with?
↓
ASK
Can AI help reveal, compare, explain or extend this pattern?
↓
JUDGE
Does the pattern suggested by AI actually fit the evidence and context?
↓
ACCOMPLISH
How can this validated understanding improve an action, decision or outcome?
So the fuller model becomes:
Human → Pattern → AI → Pattern → Judgment → Accomplishment
This may be one of the most useful ways to explain Pattern AI.
14. A Historical Parallel
The parallels between the two eras are striking.
| Earlier Computing Era | AI Era |
|---|---|
| Computers controlled largely by specialists | Advanced AI initially built by specialists |
| Personal computers expand access | Generative AI expands access |
| Graphical interfaces reduce technical barriers | Natural language reduces interaction barriers |
| Spreadsheets empower business users | AI assistants empower knowledge workers |
| End-User Computing emerges | End-User AI is emerging |
| Computer literacy becomes important | AI literacy becomes important |
| Users create models and applications | Users increasingly create analyses, content, workflows and even software |
| Need for quality controls appears | Need for verification and judgment intensifies |
There is, however, one enormous difference.
The spreadsheet mainly helped users compute.
Generative AI can participate in activities that resemble aspects of language, reasoning, creation and knowledge work.
Therefore End-User AI may eventually become even more consequential than End-User Computing.
15. From GUI to LUI
Another way to understand the transition is through interfaces.
Computing progressively moved through interfaces such as:
Command Line Interface
↓
Graphical User Interface
↓
Touch Interface
↓
Conversational / Language Interface
The GUI helped people operate computers without remembering commands.
The language interface goes further.
It lets people describe what they want in something close to ordinary human language.
This means the interface itself increasingly resembles conversation.
But conversational simplicity creates a paradox:
AI becomes easier to operate while becoming harder to judge.
Almost anyone can ask an AI system a question.
Far fewer users can reliably determine whether the answer is complete, appropriately reasoned, biased, fabricated, oversimplified or unsuitable for the situation.
Therefore the decisive End-User AI skill may not ultimately be prompting.
It may be judgment.
16. Pattern Literacy Could Become Part of AI Literacy
If Pattern AI is developed further, Pattern Literacy could become one component of broader AI literacy.
Pattern Literacy can be described as the ability to:
SEE → DESCRIBE → QUESTION → COMPARE → TRANSFER → JUDGE patterns.
A learner might progress through:
Pattern Curiosity
I notice that there may be a recurring structure.
↓
Pattern Fluency
I can describe, compare and work with patterns.
↓
Pattern Instinct
I naturally search for underlying patterns when confronting unfamiliar problems.
This capability is useful even without AI.
With AI, however, it becomes particularly powerful because the user and the machine can effectively meet around patterns.
The machine operates through learned patterns.
The human interprets patterns through meaning and context.
The collaboration happens between the two.
17. The End-User AI Equation
We can therefore express End-User AI very simply:
Domain Knowledge + Pattern Thinking + AI Collaboration + Human Judgment = End-User AI Capability
Notice what is absent.
The equation does not require everyone to become an AI engineer.
That does not mean technical AI education becomes unnecessary.
Technical specialists remain essential.
Instead, the argument is that society now needs two complementary educational tracks.
Build AI
For people developing AI systems:
Mathematics → Data → Algorithms → ML → Models → Engineering → Deployment
Use AI
For people applying AI:
Domain → Problem → Pattern → AI Interaction → Verification → Judgment → Accomplishment
Pattern AI belongs primarily in the second track while providing a conceptual bridge to the first.
18. Why This Matters for Education
This distinction could have substantial consequences for AI education.
Today, introductory AI education often begins with:
What is AI?
then:
Machine Learning
then:
Algorithms
then perhaps:
Python
That route makes sense for future AI engineers.
But imagine a first-year engineering student, medical student, management student, teacher or working professional.
Their immediate question may be different:
How does AI relate to what I already learn and do?
An End-User AI curriculum could begin with:
Your Domain
↓
Your Problems
↓
Patterns in Those Problems
↓
How AI Works with Patterns
↓
Working with AI
↓
Judging AI
↓
Applying AI
Only then does the learner go deeper into technical AI according to need.
This provides a different entry point, not a replacement for technical education.
19. Why This Matters for Work
The International Labour Organization's work on generative AI and employment increasingly frames the issue around occupational exposure and the transformation of tasks, rather than assuming that entire occupations simply disappear. Its 2025 work provides a refined global framework for understanding different levels of occupational exposure to generative AI. (International Labour Organization)
This distinction between jobs and tasks is important.
A profession is made of many tasks.
AI may automate some.
Augment others.
Transform still others.
And leave some largely human.
The practical professional question therefore becomes:
Which parts of my workflow should involve AI, and which require my own expertise and judgment?
That is another End-User AI question.
The professional of the future may not necessarily be:
Human OR AI
but increasingly:
Human + AI
with the human responsible for context, goals, judgment and accountability.
20. The Real End User Is a Domain Expert
This is perhaps the strongest lesson from End-User Computing.
The accountant using a spreadsheet was not an inferior programmer.
The accountant possessed something the programmer often did not:
deep knowledge of the accounting problem.
Similarly, the teacher, doctor, lawyer, engineer, manager, entrepreneur and student bring something essential to AI:
domain context.
AI may possess enormous generalized capability.
But usefulness emerges when that capability meets a real problem in a real context.
Therefore End-User AI should not be framed as:
“Teaching ordinary people a little AI.”
It should instead be framed as:
Giving people the capability to bring AI intelligently into domains they already understand—or are learning to understand.
That is a much more empowering proposition.
21. But End-User AI Also Recreates an Old Risk
History gives us a warning.
End-User Computing empowered users enormously, but end-user-created systems also raised concerns involving errors, maintainability, security, verification and quality. Research on End-User Development eventually expanded toward End-User Software Engineering precisely because empowering people to build computational artifacts also created responsibility for their reliability. (Wikipedia)
AI repeats this problem at a new level.
The easier AI becomes to use, the easier it becomes to generate:
plausible misinformation,
weak analysis,
incorrect code,
fabricated references,
poor decisions,
biased interpretations,
and confident-looking errors.
The solution is not to prevent end users from using AI.
The lesson from computing history suggests something else:
Empowerment must grow together with competence and controls.
For End-User AI, that means AI literacy, domain knowledge, verification and judgment.
22. Pattern AI Should Therefore Be About Agency, Not Automation
This is an important philosophical boundary.
The ultimate objective of Pattern AI should not be:
Get AI to do everything for me.
It should be:
Help me understand enough to accomplish more intelligently with AI.
That means AI can:
Reveal patterns.
Explain patterns.
Compare patterns.
Generate possible patterns.
Transfer patterns between domains.
Challenge our interpretation of patterns.
But the human remains responsible for:
purpose, context, values, verification, judgment and action.
That makes Pattern AI fundamentally human-centred.
23. A Possible New Map of the AI Landscape
We can now construct a useful conceptual map:
Layer 1 — AI Science
How can machines exhibit intelligent capabilities?
Layer 2 — AI Engineering
How do we build and deploy reliable AI systems?
Layer 3 — AI Products
How do we package AI capabilities into useful applications?
Layer 4 — End-User AI
How do people use AI effectively in real contexts?
Layer 5 — Pattern AI
How can people understand, interact with and judge AI through patterns?
Layer 6 — Accomplishment
What meaningful human outcome results?
The complete movement is:
AI Capability → AI Application → End User → Pattern Thinking → Judgment → Accomplishment
This moves the conversation away from technology alone.
It returns AI to its ultimate purpose:
helping people accomplish worthwhile things.
24. End-User Computing Democratized Computing. End-User AI May Democratize Intelligence Assistance.
This may be the larger historical story.
Mainframes concentrated computing capability.
Personal computers distributed it.
End-User Computing allowed millions of people to directly apply computing to their own problems.
Cloud and mobile computing made computational resources almost ubiquitous.
Generative AI now introduces another possibility:
ubiquitous intelligence assistance.
But intelligence assistance without human understanding can become dependency.
Intelligence assistance with judgment can become augmentation.
That distinction may define the next phase of AI adoption.
25. The Pattern AI Proposition
The proposition can therefore be stated clearly:
Pattern AI is an end-user approach to Artificial Intelligence that helps people recognize the patterns in their problems, collaborate with AI around those patterns, critically judge AI-generated patterns and convert the resulting understanding into accomplishment.
Its purpose is not to teach everyone to build AI.
Its purpose is to help everyone become more intelligent in the way they use AI.
And its operating discipline can remain remarkably simple:
THINK → ASK → JUDGE → ACCOMPLISH
with Pattern Thinking running through the entire cycle.
26. From “How Does AI Work?” to “How Do We Work With AI?”
For decades, an important question was:
How do we make computers easier for people to use?
End-User Computing was one answer.
Today the question is changing:
How do we make humans capable of working intelligently with AI?
The answer cannot simply be better interfaces.
It cannot simply be better prompts.
And it cannot simply be more powerful models.
We need better human frameworks.
Frameworks for understanding.
Frameworks for questioning.
Frameworks for recognizing patterns.
Frameworks for verification.
Frameworks for judgment.
Frameworks for accomplishment.
That is the space in which Pattern AI can develop.
Conclusion: A New End-User Era
The history of computing repeatedly shows a movement of capability toward the user.
Computing moved from specialists to desktops.
Software creation moved partly from programmers toward end-user development.
Information moved from institutional databases toward web search.
Computing moved from desktops into pockets.
And now artificial intelligence is moving from laboratories and specialized systems directly into everyday human workflows.
We may therefore be entering the era of:
End-User AI
If so, the challenge is no longer merely making AI available.
The deeper challenge is making humans capable users of AI.
That requires more than prompting.
It requires the ability to understand problems, recognize structures, collaborate with machines, question outputs, verify claims and exercise judgment.
This is where the historical lesson of End-User Computing meets the emerging idea of Pattern AI.
End-User Computing put computing power into the hands of the user.
End-User AI is putting intelligence assistance into the hands of the user.
The next challenge is to ensure that the user's own intelligence remains firmly in the loop.
And that leads to a concise positioning for Pattern AI:
Pattern AI — A Human Thinking Framework for End-User AI
See the Pattern. Think. Ask. Judge. Accomplish.
Source Note & Fact-Check
The historical claims in this article are grounded in established End-User Computing and End-User Development literature, including Rockart and Flannery's 1983 work on managing end-user computing, ACM research on End-User Development, and Bonnie Nardi's research on End-User Programming. (ACM Digital Library)
The connection between End-User Development and AI is supported by a 2023 systematic literature review specifically examining End-User Development for Artificial Intelligence, alongside newer research exploring generative-AI-assisted End-User Development. (arXiv)
Claims concerning AI literacy are supported by UNESCO and OECD frameworks emphasizing responsible use, human-centred thinking, critical assessment and broader AI competencies rather than prompting alone. (UNESCO)
Workplace evidence comes from NBER research on generative AI in customer support and Harvard/BCG research examining knowledge-worker performance and the “jagged technological frontier.” (National Bureau of Economic Research)
Important conceptual qualification: “End-User AI” is used here as an emerging conceptual category and proposed framing. It should not be represented as a newly invented term, nor as a universally established discipline with one accepted definition. Existing scholarship already discusses End-User Development for AI and related ideas. The novel proposition advanced here is the positioning of Pattern AI as a human thinking framework for the End-User AI layer.
