| |
|
| |
Your AI agent runs for days on Temporal. It retries, waits, resumes across failures, reliably. But when it's about to do something irreversible, who approves it, & on what surface?
Durable execution engines make agents survive; they don't give a human anything to see or sign off on. This talk is about building that face: human-in-the-loop for long-running agents on Temporal.
Approval gates as signals the workflow waits on, a live view of the running graph, & an exportable decision trail - engine below, humans on top. Architecture, the signal/timer plumbing, & a live demo, including what broke.
This event is a part of #SFTechWeek-a week of events hosted by VCs & startups to bring together the tech ecosystem. Learn more at www.tech-week.com
Key takeaways
* Human-in-the-loop is not a UI afterthought - it maps cleanly onto durable-execution primitives (signals/timers), so accountability becomes part of the workflow, not a bolt-on.
* The surface a human approves on must show workflow context, not engine telemetry - render the decision on the same graph that executes.
* The interface layer is engine-agnostic: reliability comes from the engine (Temporal), accountability from the layer on top.
ukasz Jawa is the CTO of Synergy Codes, a workflow & diagramming agency from Wrocaw, Poland - the team behind Workflow Builder. He leads the product's technical direction & its work with AI teams: putting a human approval step in front of agents that run for days on durable-execution engines like Temporal.
Outside work he is into physics, history, & hiking.
|
|
|
|
|
|
|
|
