Both of these break a large task into smaller ones. Which one do I want?
Whether the breakdown already exists. HTN executes a decomposition somebody encoded in advance as domain methods. Goal decomposition produces a decomposition for an objective that arrived vague.
Hierarchical Task Network (HTN) Planning
You know how this work splits up, the split is stable, and you want the same structure applied reliably every time.
Intelligent Goal Decomposition
Nobody has agreed what done means yet, and the first useful output is a set of sub-goals with success criteria attached.
What choosing wrong costs
Reaching for HTN without the domain methods means writing the domain methods, and that authoring effort is the entire cost of HTN, moved to the start of your project where it is least visible. Reaching for goal decomposition when the structure is already known trades away repeatability for a flexibility nothing needed.
Both route work to a more specialised agent. What is the difference?
Who holds control afterwards. A supervisor delegates and keeps it, fanning work out and collecting results back. A handoff gives control away and does not expect it back.
Supervisor-Worker Pattern
The subtasks are independent, you want them running at once, and something has to reconcile the results into one answer.
Handoff Orchestration
One specialist should own the rest of the interaction, because splitting it would split the context it depends on.
What choosing wrong costs
Handoff applied to parallelisable work serialises it, and you pay for the coordination without the speedup. Supervisor applied to a conversation that wanted a single owner leaves the user talking to a coordinator that re-delegates every turn and loses the thread between them.
These two names look interchangeable. Are they the same thing?
Who opens the exchange, and whether the medium is fixed. Chat is a message log the user drives, and its problems are threading, context and rich content. Conversational covers an agent that may speak first and may not be speaking in text at all.
Chat Interface Patterns
The surface is a transcript the user scrolls, and the hard parts are branching topics and keeping earlier context reachable.
Conversational Interface Patterns
The agent initiates on context, or the same interaction moves between text, voice and visual.
What choosing wrong costs
Building thread branching for a voice agent that never renders a transcript is machinery nobody sees. Building proactive multimodal orchestration for what is really a support window is a large surface area to maintain around a text box.
In the loop, on the loop. Is there a real difference or is it jargon?
Whether the system waits. In-the-loop blocks: nothing proceeds until a person decides. On-the-loop runs at full speed and a person watches, stepping in when something trips an alert.
Human-in-the-Loop
A wrong action is unacceptable and the latency of waiting for a human is affordable.
Human On the Loop
Throughput matters, failures are rare, and a person can realistically reach the controls before the damage lands.
What choosing wrong costs
In-the-loop on a high-volume path becomes a queue nobody services, and the approvals get rubber-stamped. That is worse than having no gate, because it manufactures an audit trail of decisions nobody examined. On-the-loop where nobody actually watches the dashboard is full autonomy wearing a compliance story.