Same topic, related formats. Practice these next.
Flat planning forces one reasoning step to juggle strategy and tactics at once, so the model loses the high-level goal while chasing low-level detail, and coherence collapses on long tasks.
Imagine planning a road trip. There is a big-picture question, which cities do I visit and in what order, and a tiny question, which exit do I take and where do I refuel. If you try to answer both in the same breath, you keep losing the thread. You stare at a gas-station sign while forgetting you meant to reach the coast by sunset. Now imagine a friend handles the overall route and you only handle the next turn. Each of you thinks about one thing at a time, and neither gets overwhelmed. An agent works the same way. Flat planning makes one mind decide the whole strategy and every tiny tool call together, which is exhausting and error prone. Hierarchical planning splits the trip into a route planner and a turn taker, so each step stays simple and the goal never slips out of view.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
Everything important, quickly.
State that the failure is context confusion from mixing abstraction levels, name the strategic, tactical, and operational levels, give the three decomposition wins of manageable steps, parallelism, and error isolation, then explain the cascade failure mode and how plan-and-execute with replanning manages the tradeoff.
Real products, models, and research that use this idea.
What an interviewer would ask next. Try answering before peeking at the approach.
Red flags and common mistakes that signal junior thinking. Click to expand.
Assuming the failure is about token count or tool budgets. The real problem is cognitive: one reasoning step working at every abstraction level at once loses the high-level goal.
The night-before-the-interview bullets. Scan these on the way to the call.
Primary sources. Skim if you want the original framing.