Mental Model
Two Tracks in Partnership
The Core Insight
Digital work flows through two parallel tracks — human and AI — that interleave and depend on each other. The human track is primarily about specification: defining intent precisely enough for systems to execute. The AI track is primarily about production: executing that specification at speed and scale.
Neither track works alone. AI can produce fast but needs human specification. Humans can specify but benefit from AI's ability to explore options and execute at speed.
Human Track (Specification)
Understanding the problem
Defining intent
Evaluating output
Accepting or redirecting
AI Track (Production)
Researching options
Proposing approaches
Implementing
Delivering
The tracks interleave constantly. Defining intent and proposing approaches happen together. Evaluating output and implementing form a feedback loop. Both tracks converge at completion.
The cost of producing software is collapsing toward zero. The cost of not knowing what to build — specifying badly, vaguely, or not at all — is compounding. The bottleneck in all knowledge work has shifted from production to specification.
HAI was built for this shift before we had a name for it. You can now build the wrong thing at unprecedented speed and scale. HAI helps you build the right thing.
“Whenever I find a bug that my agent has introduced, I realize it's more likely that my spec was not clear enough.”
Direction
What to create and why — the specification
Judgment
Evaluating tradeoffs, quality, fit
Context
Domain knowledge, relationships, history
Review
Validating AI output against intent
Accountability
Standing behind decisions and outcomes
Research
Exploring options, gathering information
Synthesis
Combining information into coherent outputs
Production
Creating artifacts (code, text, analysis)
Iteration
Rapid refinement based on feedback
Consistency
Applying patterns uniformly at scale
Work Modes — Who Leads
AI-led
AI does the work, human ensures quality
When it's right: Implementation of well-defined specs, routine tasks
Example: Adding a CRUD endpoint following existing patterns. Path is clear, AI implements, human reviews.
Collaborative
Human and AI work together iteratively
When it's right: Feature development, problem-solving, design
Example: Designing a new feature with tradeoffs to evaluate. Figuring it out together, iterative.
Human-led
Human decides, AI supports
When it's right: Strategy, client relationships, novel decisions
Example: Pricing strategy, client relationship decisions. Human drives, AI supports.
This isn't a hierarchy — different work legitimately needs different modes. As AI capabilities evolve, work naturally shifts modes. What was Collaborative in 2026 may be AI-led in 2028. This drift tells the story of progress.
Involvement Levels — Human Attention
Quick review, routine approval
Standard patterns, minimal decisions needed
Minutes of human time
Active participation, some decisions
Judgment calls required, familiar territory
An hour or more of human time
Deep thinking, significant decisions
Multiple stakeholders, important choices, novel territory
Half a day or more
The specification/production split applies everywhere. The work mode and involvement scales generalize to any digital work.
| Domain | Human Specification | AI Production |
|---|---|---|
| Content Creation | Voice, strategy, audience intent | Drafting, variations, formatting |
| Data Analysis | Questions to answer, interpretation | Processing, pattern finding, visualization |
| Research | Direction, synthesis framework | Gathering, summarizing, organizing |
| Design | Vision, constraints, feedback | Generation, iteration, variations |
| Documentation | Structure, accuracy criteria | Drafting, consistency, formatting |