Put a person in the middle of a pipeline
The run stops at the approval and waits. That is the difference between something you can put in front of a customer and a script somebody watches.
Everything below describes what this is being built to do. It needs Studio, which you can look around today but cannot yet use for the whole of this job. The problems are real now; the answers are the part still being built.
Some steps should not happen because a model was confident. They should happen because a person said so.
Why it is hard
Confidence is not authority
A pipeline that spends money, sends a message or changes a record on a model's say-so is one bad output away from doing it wrong at speed.
Bolting approval on means stopping the pipeline
Usually the pause is a separate system, so the run ends, somebody is emailed, and a second run starts with state carried by hand.
Nobody can see what is waiting
When approvals live outside the pipeline, work stalls invisibly and the first sign is somebody asking why nothing finished.
What answers it
The pause is a step
An approval node between two steps, in the same graph, drawn like everything else. The canvas supports drawing one today.
Waiting is visible where the work is
A stalled approval shows on the diagram rather than in somebody's inbox, which is the difference between a queue and a mystery.
Nothing downstream runs early
The run holds at the approval and continues from it. This needs execution, which is the part not built.
What you use from Kracht
Related jobs
- StudioplannedRun a multi-step agent pipeline on GPUs you controlEach step gets the card it needs rather than the one the whole pipeline was sized for. The shape of the work and the cost of the work become the same picture.Read the job
- StudioplannedBring a workflow you already wroteLangGraph, LangChain and CrewAI graphs are import sources rather than competitors. The intent is that an existing graph arrives as nodes you can see, not as a rewrite.Read the job
- StudioplannedSee what an agent run actually did, span by spanEvery tool call, prompt and token recorded against the step that made it, so a bad answer traces back to the exact call responsible.Read the job