Contents
The Problem
The execution spiral is the signature failure mode of AI-native work. I instruct the AI. The AI produces output. The output is close but not right. I correct. The AI adjusts. The adjustment introduces a new deviation. I correct again. Each correction narrows one gap while opening another. After several iterations, I have lost track of the original intent. Activity stays high. Output keeps flowing. But the output no longer maps to the target.
The spiral is not a failure of the AI. It is a failure of alignment between what I hold mentally and what the AI receives as instruction. The gap between those two states compounds with each iteration. Standard debugging does not help — step-through debugging examines code line by line, addressing symptom rather than cause. If my instructions were misaligned, perfect debugging of the output still produces misaligned results. Rubber duck debugging externalizes my thinking but gives me no structured response. Root cause analysis works retrospectively, but in AI-native execution, context degrades rapidly — analysis delayed is context lost.
The worst response is to push through the confusion. Every misaligned iteration compounds the divergence. The second worst response is to stop and plan from scratch — that kills the velocity the methodology exists to maintain. I needed a third option: structured recovery that restores alignment without sacrificing momentum.
What Micro-Triage Actually Is
Micro-Triage is a six-step diagnostic sequence, timeboxed to 15-30 minutes, that restores alignment between operator intent and AI execution. It deploys after the operator has recognized a spiral and paused. Each step builds on the output of the previous one — this is escalation, not repetition.
What it provides:
- Structured recovery — a repeatable protocol that turns spiral detection into corrective action within 30 minutes
- Joint diagnosis — the operator and AI diagnose together, because the gap exists between their respective states and neither can find it alone
What it does not provide:
- A substitute for upstream clarity — if Vision is unclear or the Target is unlocked, Micro-Triage will not resolve the problem because the problem is not tactical
- Unlimited diagnosis time — if 30 minutes of structured dialogue does not resolve it, continued diagnosis will not either; the mechanism escalates rather than allowing circular debugging
The hard timebox is the structural backbone. Diagnosis without constraint becomes rumination. If the problem cannot be resolved in 30 minutes of structured operator-AI dialogue, the problem is upstream. Escalation clears the state and redirects to fundamentals.
The Six-Step Sequence
Micro-Triage follows a fixed escalation path. Each step feeds the next. Skipping steps degrades the diagnosis.
Step 1: Reground — "What is the Target?" Return to the Target. State the concrete outcome in plain language. This forces me to articulate intent, which may have drifted during the spiral.
Step 2: State — Explain the task to the AI. Describe what I am trying to accomplish in the current execution thread. Not the overall project — the specific task that spiraled.
Step 3: Mirror — Ask the AI to explain what it sees. The AI describes its understanding of the task, the current state, and what it believes I want. This is the critical step. It creates shared representation between operator and AI.
Step 4: Align — Discuss the gap. Compare my intent from Steps 1-2 with the AI's understanding from Step 3. The gap between these two states is the source of the spiral.
Step 5: Diagnose — What did the AI do versus what I asked for? Identify the specific divergence. Not "it's wrong" but "I asked for X, the AI understood Y, and produced Z." Precision here prevents repeated misalignment.
Step 6: Execute — Define the corrective action and run it. Specific, concrete, testable. Not "fix the thing" but a defined correction against the diagnosed gap. If it resolves, resume at full velocity. If 30 minutes pass without resolution — stop and escalate.
The escalation protocol prevents infinite recursion. Resolved in the timebox: resume execution. Partially resolved: continue with reduced scope and flag the remaining gap. Unresolved but tactical: run a full Refresh to clear state. Unresolved and systemic: revisit Vision or Target — the problem is upstream. Triage that does not resolve within its timebox is, by definition, addressing the wrong problem.
What the Data Shows
Micro-Triage was validated through production of ten software systems totaling 596,903 lines of code between October 2025 and February 2026 (2,561 raw commits, approximately 2,246 deduplicated). As an operator behavior, Micro-Triage does not leave direct git artifacts — validation is inferential through observable commit patterns.
| Evidence Pattern | Observation |
|---|---|
| Gap-then-pivot sequences | 30-60 minute commit gaps followed by commits in different subsystem or with different approach |
| Rework clustering | 12.1% of commits are product defect fixes, clustered rather than uniformly distributed |
| Approach pivots within sessions | Same-day commits showing feature approach changes mid-session after diagnostic gaps |
| Sustained velocity | 29 commits/day average maintained across 4 months |
The 12.1% product bug rate — against industry norms of 20-50% — suggests spirals were caught early through tactical intervention rather than propagating into systemic rework. The timebox prevents extended periods of misaligned execution that would generate compound defects. If execution spirals were not resolved tactically, velocity would degrade as I spent increasing time on rework. The sustained 29 commits per active day across four months is indirect evidence that a recovery mechanism was operating effectively.
The rework clustering pattern is particularly telling. Defect fixes cluster into concentrated correction sessions rather than distributing uniformly. This is consistent with spiral detection followed by structured correction — exactly what the Reground, State, Mirror, Align, Diagnose, Execute sequence produces.
How to Apply It
1. Recognize the Spiral Early The spiral has a characteristic signature: you are correcting the AI's output, but each correction opens a new gap. Activity is high, but you have lost track of your original intent. When you notice the correction loop, stop. Do not push through. The earlier you catch it, the less misaligned context you need to unwind.
2. Run the Six Steps in Order Reground on the Target. State the specific task. Ask the AI to Mirror its understanding back to you. Compare your intent with the AI's model. Diagnose the specific divergence — "I asked for X, AI understood Y, produced Z." Then define and execute the correction. Do not skip steps. The Mirror step is where most of the diagnostic value lives — it makes the invisible alignment gap visible.
3. Enforce the Timebox Set a 30-minute hard boundary. If structured dialogue between you and the AI has not resolved the spiral in 30 minutes, the problem is not tactical. Escalate: run a full state reset if the issue is tactical but stubborn, or revisit your Vision and Target if the issue is systemic. Continued diagnosis past the boundary produces rumination, not resolution.
4. Feed Resolved Spirals Back Into Foundation Every spiral you diagnose and resolve is a pattern. Record the signature — what the spiral looked like, where the alignment gap was, what corrected it. Future spirals with similar signatures get recognized faster. The diagnostic library compounds over time, making each Micro-Triage faster than the last.
References
- Rollbar (2021). "Developer Survey: Fixing Bugs Stealing Time from Development." 26% of developers spend up to half their time on bug fixes. Source
- Coralogix (2021). "This Is What Your Developers Are Doing 75% of the Time." Developer time allocation to debugging and maintenance. Source
- Hunt, A. & Thomas, D. (1999). The Pragmatic Programmer. Addison-Wesley.
- Keating, M.G. (2026). "Vision." Stealth Labz CEM Papers. Read paper
- Keating, M.G. (2026). "Foundation." Stealth Labz CEM Papers. Read paper
- Keating, M.G. (2026). "Pendulum." Stealth Labz CEM Papers. Read paper
- Keating, M.G. (2026). "Target." Stealth Labz CEM Papers. Read paper