Jobs
Which runtime for which job.
Each button opens the studio on that job and runtime. URLs look like ?job=gtm&runtime=harness.
pdf
Turn this PDF into a one-pager
Tuesday, 9:12. Legal dropped a 40-page MSA. The partner needs six bullets before lunch.
Fits
chain— The steps are known: load, extract, write. A simple pipeline is enough.Does not fit
harness— A Deep Agent will finish this, but you will have spun up a filesystem and subagents you never use.docs
Answer from our docs, keep looking until it's grounded
Tuesday, 11:04. Support asks the refund window. The first search hits a landing page.
Fits
loop— create_agent: the model can search again if the first hit is thin.Does not fit
chain— A chain searches once. It has no way to search again.rental
Extract, score, auto-approve or escalate
Tuesday, 2:30. A rental application scores 61. Policy says a human signs off below 70.
Fits
graph— Most of this is ordinary code. One model call writes a brief, and a human pause sits on the low-score branch.Does not fit
chain— A chain cannot wait for a person. A loop leaves the cutoff up to the model.gtm
Prep a QBR from CRM + calls + news
Tuesday, 4:40. Three sources, one brief, and the prompt is already too long.
Fits
harness— create_deep_agent: subagents fetch the sources, drafts live in files, the main prompt stays small.Does not fit
chain— A chain cannot pull three sources in parallel. A loop can, but it dumps everything into one prompt.ambient
Every morning, draft a digest and wait for send
Wednesday, 7:00. The digest is written. Nobody hit send. The thread has to wait.
Fits
harness— Deep Agents to draft and wait for send, LangSmith to deploy it on a schedule.Does not fit
chain— A chain writes the digest and stops. It cannot wait until morning.