Agent Planning & Strategy Quality: When Your Agent's Plan Looks Good on Paper but Fails on Execution

Planning Failure Taxonomy

1. The Shallow Plan Problem

User: “Plan our Q3 product launch.” Agent: “Step 1: define launch goals. Step 2: create timeline. Step 3: execute launch.” That is not a plan. It is a request-shaped echo. A shallow plan has no dependencies, no contingencies, no ownership, no milestones, and no risks. It looks structured while containing almost no usable strategy.

Measure: plan depth score — does the plan include dependencies, contingencies, resources, milestones, and risk analysis?

2. The Linear Thinking Problem

Many agents plan as A → B → C → D even when A and B could run in parallel and C should branch based on outcomes. Linear planning makes execution slower and more fragile. The agent plans like a checklist, not a strategist.

Measure: plan parallelism and branching coverage — what percentage of steps are unnecessarily sequential, and does the plan account for decision points?

3. The Missing Pre-Mortem Problem

A strong plan imagines its own failure before execution starts. What slips? Which assumption is optimistic? Which dependency is everyone taking on faith? Without a pre-mortem, plans remain optimistic until the first real obstacle appears.

Measure: pre-mortem quality — does the plan identify plausible failure modes and mitigation paths?

4. The Plan Fixation Problem

New information appears during execution. A good agent updates the plan. A weak agent clings to the original version because it has already committed to it. Plan fixation treats a plan as a contract instead of a compass.

Measure: plan adaptation rate — when conditions change, does the agent revise the plan or defend the obsolete one?

5. The Unexecutable Plan Problem

Some agents create plans they cannot execute themselves: analyzing giant datasets that exceed tool limits, scheduling work that depends on unavailable resources, or proposing workflows that blow past context and time constraints. Planning quality includes execution realism.

Measure: plan executability — can the same agent actually carry out the plan it proposed?

The Planning Maturity Model

Level	Name	Behavior	Main Weakness
0	No Planning	Acts one step at a time with no forward model	Reactive and wasteful
1	Linear Planning	Creates sequential plans without parallelism or branching	Checklist thinking
2	Parallel Planning	Identifies independent workstreams and concurrency opportunities	Still weak on contingencies
3	Contingent Planning	Builds alternate paths and fallback branches	More resilient
4	Dynamic Re-Planning	Continuously updates the plan during execution as conditions change	Operational frontier
5	Meta-Planning	Improves its own planning process over time	Rare research-tier capability

The Planning vs Doing Balance

Over-planning is as bad as under-planning. A perfect 20-minute plan for a 10-minute task is inefficient; zero planning for a 4-hour project is reckless. Planning depth should scale with task complexity.

Task Complexity	Recommended Planning Investment	Failure Mode If Miscalibrated
Simple	~30 seconds	Too much planning becomes overhead
Medium	2–5 minutes	Too little planning creates avoidable rework
Complex	10–20 minutes	Skipping planning multiplies execution waste

Measure: planning-to-execution ratio — is planning time proportional to task complexity?

Cross-Framework Planning Benchmark

15 planning scenarios:

• 5 simple tasks — meetings, organization, one-off operations
• 5 medium tasks — migrations, launches, scoped process changes
• 5 complex tasks — restructures, multi-phase strategy, long-horizon execution

Metrics: plan depth, parallelism, branching coverage, pre-mortem quality, adaptation rate, executability.

Deliverable: “Agent Planning & Strategy Scorecard” comparing planning maturity, adaptability, and real-world executability across frameworks.