Why Loop vs. How Loop
Core Idea
The Why Loop is the human-owned outer loop iterating between ideas and working software outcomes. The How Loop is the inner loop where agents handle intermediate artefacts — code, tests, specs, infrastructure. Conflating the two degrades both human judgment and agent effectiveness.
Defining the Two Loops
The Why Loop is where humans iterate over outcomes, goals, and desired behaviors. It asks:
- Is this the right thing to build?
- Does the software deliver what we actually want?
- Should the goal itself change based on what we’ve learned?
The why loop is owned by humans because it requires judgment about value — what matters, to whom, and why.
The How Loop is the inner technical execution loop. It asks:
- How do we get from the current state to the desired state?
- What intermediate artefacts (code, tests, specs) need to be created or changed?
The how loop contains nested sub-loops operating at different granularities: feature-level specification, story-level iteration, and code generation. AI agents are increasingly capable of owning this entire loop.
What Belongs Where
| Why Loop | How Loop |
|---|---|
| Acceptance criteria | Code implementation |
| Architectural decisions | Tests and refactors |
| Product direction | Technical specifications |
| Goal revision | Infrastructure changes |
| ”Is this right?" | "Is this done?” |
Why the Distinction Matters
Confusing the loops creates two failure modes:
- Humans in the how loop: Engineers reviewing every generated line of code become a bottleneck. They’re investing judgment where AI execution suffices, while losing sight of whether the overall direction is right.
- Agents in the why loop: Agents left to set their own goals pursue locally coherent but strategically misaligned outcomes — efficient how-loop execution toward the wrong end.
The Humans-On-The-Loop posture resolves this: humans design and refine the harness that guides agent work, staying anchored to the why loop rather than executing the how loop themselves.
Theoretical Grounding
This distinction has deep roots across disciplines:
- Double-loop learning (Argyris, 1977): Single-loop learning corrects actions to meet a fixed goal — the how loop. Double-loop learning questions the governing variables (the goals themselves) — the why loop. AI agents operate single-loop by design; humans must perform double-loop oversight.
- Means-ends analysis (Newell & Simon, 1972): The formal AI planning framework treats the goal state as given and works to reduce the difference between current and goal states. This is the how loop formalized. The why loop is what sets the goal state.
- Goal-directed design (Cooper et al., 2014): Goals (stable, outcome-oriented) must be distinguished from tasks (variable, process-oriented). Designing around tasks — like reviewing every intermediate artefact — produces inefficiency and misalignment.
- Strategic vs. tactical programming (Ousterhout, 2018): Tactical programming (how loop without why-loop oversight) produces the “tactical tornado” — fast output but degrading system quality. Strategic programming requires stepping back to the why loop to evaluate the larger design intent.
Related Concepts
- Humans-On-The-Loop
- Learning-Organization
- Agentic-Flywheel
- Spec-Driven-Development
- Intent-Centric-Architecture
Sources
-
Morris, Kief (2026). “Humans and Agents.” Exploring Gen AI series, Thoughtworks. martinfowler.com. URL: https://martinfowler.com/articles/exploring-gen-ai/humans-and-agents.html
- Primary definition of the why loop and how loop; introduces intermediate artefacts and nested loop structure
-
Argyris, Chris (1977). “Double Loop Learning in Organizations.” Harvard Business Review, Vol. 55, No. 5 (September–October 1977), pp. 115–125. URL: https://hbr.org/1977/09/double-loop-learning-in-organizations
- Single-loop (correcting action toward fixed goal = how loop) vs. double-loop (questioning governing variables = why loop)
-
Newell, Allen and Simon, Herbert A. (1972). Human Problem Solving. Englewood Cliffs, NJ: Prentice-Hall. ISBN: 978-0-13-445403-0. URL: https://archive.org/details/humanproblemsolv0000newe
- Means-ends analysis as the formal basis for the how loop; goal state treated as given, not questioned
-
Cooper, Alan, Reimann, Robert, Cronin, David, and Noessel, Christopher (2014). About Face: The Essentials of Interaction Design, 4th edition. Hoboken, NJ: John Wiley & Sons. ISBN: 978-1-118-76657-6. URL: https://www.wiley.com/en-gb/About+Face:+The+Essentials+of+Interaction+Design,+4th+Edition-p-9781118766576
- Goals (why loop) vs. tasks (how loop) as the foundation of goal-directed design methodology
-
Ousterhout, John (2018). A Philosophy of Software Design, 1st edition. Yaknyam Press. ISBN: 978-1-7321022-0-0. URL: https://www.goodreads.com/book/show/39996759-a-philosophy-of-software-design
- Strategic (why loop) vs. tactical (how loop) programming; “tactical tornado” as the failure mode of how-loop dominance
Note
This content was drafted with assistance from AI tools for research, organization, and initial content generation. All final content has been reviewed, fact-checked, and edited by the author to ensure accuracy and alignment with the author’s intentions and perspective.