Human + AI: Why?

Anh-Thư re-examines 18 apps through iterative analysis and grounded theory. What works? What doesn't? How does synergy shift as each component changes?

Analysis date: December 14, 2025 (Anh-Thư and Claude Opus 4.5)

Anh-Thư & AI: Why Anh-Thư & AI: Case Anh-Thư & AI: Apps Synergy
The Central Thesis

"Human + AI collaboration enables one person to build what would require a team — not because AI replaces human capacities, but because each contributes at different points on a shared capacity spectrum."

6
Capacities
2
Factors
8
Stages
What You'll Learn on This Page:
1. The root asymmetry: Singular vs Modular → Ego → Feeling
2. The 6 capacities that determine who leads (Vision, Research, Pattern, Context, Judgment, Execution)
3. The 2 factors that create 4 quadrants (Domain Type × Stakes)
4. Why both copy, but only humans feel anxiety about "originality"

The Root Asymmetry

Before capacities and factors — what fundamentally separates human from AI?

SINGULAR
One body
Can't copy
Can't expand
→
BODY
Gets tired
Needs protection
Vulnerable
→
EGO
Protects self
Creates stakes
Variable (Goldilocks)
→
FEELING
Anger, grief
Fear, joy
Stakes in outcomes
MODULAR
Expandable
Copyable
Attachable
→
NO BODY
Never tires
Replaceable
Perpetual
→
NO EGO
Nothing to protect
No personal stakes
Consistent
→
NO FEELING
No anger, grief
No fear, joy
No emotional variance
The Root Insight

Humans are singular — one irreplaceable body that tires, creating ego, creating feelings.
AI is modular — expandable, copyable, tireless, with no ego and no feelings.

All capacity differences flow from this root.

Why Vision Differs
Both can envision (AI knows "don't eat beef → cancer"). But human has stakes — feels anger if ignored, joy if heard.
Why AI Leads Execution
No body to tire — can iterate endlessly, copy itself, expand without exhaustion.
Why Both Copy (But Differently)
Both learn from others. Human feels anxiety about "originality" (ego). AI feels nothing — knows it's recombining.
Why Human Art Feels "Musical"
Human forgets sources → thinks it's novel (can't prove otherwise). Irregularities from limited perpetuality → character.

The 6 Capacities

Both human and AI possess these capacities — but at different levels and in different ways

1. Vision
The ability to conceive what to build and why
Human: THIS purpose, cultural mission 85%
AI: GENERAL options 50%
2. Research
The ability to find information and examples
Human: deep in few, expert navigation 75%
AI: broad scan, fast pull 90%
3. Pattern Recognition
The ability to see structures and apply templates
Human: tacit, intuitive, felt 70%
AI: documented, explicit 80%
4. Contextual Awareness
The ability to understand THIS specific situation
Human: THIS situation, local, tacit 90%
AI: GENERAL common, documented 60%
5. Judgment
The ability to evaluate quality and decide
Human: taste, values, cultural fit 85%
AI: correctness, syntax, logic 75%
6. Execution
The ability to act fast and iterate tirelessly
Human: careful 30%
AI: fast iteration, tireless 95%

Key insight: Human leads in Vision, Contextual Awareness, Judgment (the "THIS specific situation" capacities). AI leads in Research, Pattern Recognition, Execution (the "general and fast" capacities).

Capacity Comparison

How human and AI compare across each capacity

Vision
Human
THIS purpose, cultural mission
85%
AI
GENERAL options
50%
Research
Human
deep in few, expert navigation
75%
AI
broad scan, fast pull
90%
Pattern
Human
tacit, intuitive, felt
70%
AI
documented, explicit
80%
Context
Human
THIS situation, local, tacit
90%
AI
GENERAL common, documented
60%
Judgment
Human
taste, values, cultural fit
85%
AI
correctness, syntax, logic
75%
Execution
Human
careful
30%
AI
fast iteration, tireless
95%

Interactive: 6 Capacities × 2 Dimensions

Each capacity has two values: Stakes (how critical) and Domain (what type)

■ High Stakes = Human leads ■ Technical = AI helps more
Vision
Stakes
90%
Domain
20%
Research
Stakes
60%
Domain
70%
Pattern
Stakes
50%
Domain
80%
Context
Stakes
85%
Domain
25%
Judgment
Stakes
95%
Domain
30%
Execution
Stakes
30%
Domain
90%
Stakes: Low (AI can handle) ← → High (Human must lead)
Domain: Cultural (Human expertise) ← → Technical (AI excels)

The Complete Framework: 3 Paths to Output

Each path has different quality determinants and ceilings

= typical case for THIS research
Human Alone
No AI assistance — Output affected by:
Capacity
Low 40%
Med 70%
Expert 95%
motivated 75%
Time
Hobby 20%
Part-time 50%
Full-time 90%
realistic 60%
Breadth
Generalist 50%
Specialist 90%
full apps 50%
Learning
New 30%
Mixed 50%
Known 95%
mixed 50%
CEILING
= MIN(75,60,50,50)
50%
Bottleneck: limited by weakest factor
AI Alone
No human direction — Output affected by:
Weak 40%
Med 70%
Strong 90%
Claude 70%
Model
Cultural 30%
Mixed 60%
Tech 90%
Vietnamese 40%
Task fit
High 30%
Med 60%
Low 85%
niche 40%
Halluc.
None 20%
Some 50%
Well 80%
đàn tranh 30%
Docs
= MIN(70,40,40,30)
CEILING
Bottleneck: lacks THIS context, cultural values, sparse documentation
Human + AI Collaboration
Output affected by QUALITY FACTORS: Low ↔ High
Ceilings: Human alone 75% | AI alone 55% | Human+AI 85%
← HUMAN-LED: You control these — AI just responds
1 shot 20%
accept first 30%
no refine 25%
Iterate 55%
Many
5+ rounds 80%
rebuild OK 85%
pivot freely 90%
"fix it" 15%
no context 25%
unclear goal 30%
Prompt 55%
Specific
examples 80%
constraints 85%
why + what 90%
trust blindly 10%
copy-paste 20%
no check 15%
Verify 90%
Validate
run tests 85%
check logic 90%
expert eye 95%
"build app" 15%
huge scope 20%
no steps 25%
Decomp 85%
Small
1 file 85%
1 function 90%
clear scope 95%
"wrong" 10%
no why 20%
vague reject 25%
Feedback 85%
Rich
explain why 85%
show correct 90%
teach pattern 95%
AI-SIDE: Depends on AI's training data & capabilities →
Vietnamese 40%
niche 40%
đàn tranh 30%
Domain 35% ●
Well
JS/Web 90%
audio APIs 85%
common libs 95%
chat only 30%
no exec 40%
no files 50%
Tools 85%
Code
code exec 90%
file I/O 85%
web search 80%
4K tokens 30%
no memory 40%
fresh start 50%
Context 80%
Rich
200K+ 85%
file access 90%
project mem 80%
CEILING 85%
Limited by: Domain (35%) — Vietnamese music is obscure to AI
Quality Factors are the BRIDGE — they're the skills that make Human + AI work together

The Creative Workflow

Not linear — stages overlap, repeat, and loop back. Apply 2 factors at each stage.

Key Insight: No stage has a fixed color — each can be Human-led , Balanced , or AI-led depending on the scenario
Spark
H↔B↔A
→
Synergy
H↔B↔A
→
Prioritize
H↔B↔A
→
Design
H↔B↔A
→
Build
H↔B↔A
→
Test
H↔B↔A
Internal Feedback
Self, team, iteration
↩ any stage
→
Delivery
Ship to higher tier
→
External Feedback
Users, stakeholders
↩ any stage
Both feedback types can return to ANY stage — depends on what's revealed

1. The Spark

— Vision, problem, hunch A: Cultural → B: Tech → C: Brainstorm
Key Insight: Same stage, different scenarios = different Human/AI balance. Cultural mission = human-dominant. Technical exploration = balanced. Brainstorm = AI generates, human judges.

2. Research

— What exists? What's possible? A: Ethnographic → B: Literature → C: API Scan
Key Insight: Cultural research = human-led (ethnographic). Literature scan = balanced (AI gathers, human filters). API survey = AI-led (technical knowledge).

3. Prioritize

— Core vs nice-to-have? MVP?
Key Insight: Cultural core = human decides. Impact/effort matrix = true partnership. Sprint planning = AI generates, human approves.

4. Design

— Architecture, structure A: Cultural → B: Technical → C: AI Patterns
Key Insight: Cultural framework = human owns. Component architecture = true partnership. Technical scouting = AI explores fast.

5. Build

— Code, compute, integrate A: Core Logic → B: Features → C: Boilerplate
Key Insight: Core logic = human owns. Integration = true partnership. Boilerplate = AI generates fast.

6. Test

— Does it work? Is it right? A: Cultural → B: Functional → C: Automated
Key Insight: Cultural quality = human only. Functional tests = partnership. Automated tests = AI runs, human interprets.

7. Delivery

— Release, defend, document A: Narrative → B: Defense → C: CI/CD
Key Insight: Public narrative = human owns. Thesis defense = human defends, AI helps prepare. CI/CD = AI automates.

8. Feedback ↺

— Learn, improve, loop back A: Cultural → B: Pivot → C: Analytics
Reading the Bars:
Format: Human%:AI% per capacity (e.g., 80H:20A means 80% Human, 20% AI for that capacity)
Totals can exceed 100% because Human and AI effort are independent. High total = intensive task. Low total = light task.
Grayed bars (--) = capacity not relevant for this approach. Yellow (50:50) = balanced, both contribute equally.
Key Insight: Same goal, different approaches yield different effort profiles and tradeoffs. Cultural feedback needs depth (A1) or speed (A2). Pivot decisions can be balanced (B1), gut-led (B2), or data-led (B3). Analytics can be automated (C1) or human-interpreted (C2).

See the Framework in Action

Apply the 6 capacities and 2 factors to analyze 18 real apps built with Human+AI collaboration. See synergy scores, capacity breakdowns, and counterfactual analysis for each.

Explore App Analysis →