AI Planning

Failure should be boring Approval workflows are supposed to be boring. A client submits documents, a system checks the required conditions, and an approval either happens or does not happen. Boring is good. Boring means the process does not accidentally approve a case while also escalating it as problematic. The trouble begins when a workflow is written as a best-effort model of reality. Someone encodes the actions. Someone else adds an exception. A third person adds a shortcut because the quarterly dashboard prefers speed over philosophy. Eventually, a sequence exists that should not exist. It does not look like a bug when inspected locally. Each action seems defensible. The path as a whole is the problem. ...

Schedules break in the small gaps. A delivery truck leaves at the right time, but the loading dock was not open yet. A watering pump arrives near the plant, but the tap is not being opened by the second worker at the same moment. A rescue boat reaches the correct coordinate, but after the deadline. In normal business language, these are “coordination issues.” In automated planning language, they are temporal constraints, numeric resources, durative actions, invariants, and interference rules. ...

Schedule three people, one car, two children, five afternoon activities, and several goals that quietly hate each other. Then ask a normal person to find the best plan. That is already a planning problem. Now ask the same person to understand why a plan failed, which goals caused the failure, what could be added without breaking the plan, and what must be sacrificed if one more constraint is enforced. ...

Plans fail in painfully ordinary ways. A warehouse robot cannot both reserve the last pallet slot and keep the aisle clear. A field-service schedule cannot satisfy every customer window after one technician calls in sick. A compliance workflow cannot approve a transaction before the missing document exists, no matter how passionately the dashboard insists on “urgent priority.” ...

“Please double-check your work” is one of the least expensive quality-control systems ever invented. It is also one of the least dependable. A person who overlooked a constraint the first time may overlook it again. A language model is no different, except that it can produce a longer and more persuasive explanation of why the overlooked constraint was never important. ...

Robots do not move through warehouses as clean little dots on a grid. They rotate. They accelerate. They wait behind other robots. They lose time in corners. They obey controllers, not PowerPoint arrows. This is the small operational fact that makes a large amount of path-planning optimism look slightly overdressed. Multi-Agent Path Finding, or MAPF, usually asks a neat question: given many agents, each with a start and goal location, can we find collision-free paths for all of them? In the standard version, the world is a graph, time advances in discrete steps, and each robot either moves to a neighboring vertex or waits. It is elegant, measurable, and algorithmically productive. It is also not how a differential-drive robot actually behaves when squeezed through a congested warehouse aisle. ...

TL;DR for operators A new benchmark does not say that LLMs are hopeless at planning. That would be too easy, and also false. It says something more useful: frontier models are now strong enough to solve many formal planning tasks, but their competence still weakens when the task stops giving them semantically meaningful labels.1 ...

A planner stalls. Not because the goal vanished. Not because the system lacks compute. Not even because the heuristic is completely wrong. It stalls because the heuristic has temporarily stopped saying anything useful. Every nearby state looks equally unpromising, or worse, misleadingly unpromising. The algorithm is still running, naturally. It is very busy being lost. ...

TL;DR for operators When an AI system has to execute a multi-step operational plan, the tempting move is to ask the LLM for the plan. This paper argues for a less glamorous and more useful pattern: let the LLM help shrink the search problem, then let a classical planner verify and compose the actual action sequence.1 ...

TL;DR for operators SimuRA is an agent architecture that asks a simple operational question: before an AI agent clicks, searches, filters, submits, or replies, can it cheaply rehearse what might happen next?1 Not in a poetic “the machine imagines” sense, please calm down. In a practical sense: generate candidate actions, simulate their likely outcomes in a compact internal state, score those futures against the goal, and only then execute the first concrete action. ...