# Every step you take, every call you make: the reliable agent stack — Giselle van Dongen, Restate

Source: [AI Engineer](https://www.youtube.com/watch?v=cI7zfqusmFU)  
Feed7 permalink: https://feed7.dev/p/every-step-you-take-every-call-you-make-the-reliable-agent-stack-giselle-1ojh38b  
Published: 2026-09-14T15:30:30.000Z  
Trust: Source Linked (source_linked)

## Why Included

Long-running agents need durable state, retries, cancellation, and human approval that survive deploys. Restate demonstrates these as infrastructure concerns rather than prompt logic.

## Source Summary

Restate journals agent events so a failed run can resume without replaying completed work. Its demo combines a planner, human approval, parallel researchers, retries, cancellation, and session-isolated state; suspended approval waits consume **no serverless execution time**.

## Practical Implication

Treat production agents as persistent distributed processes. Put tool calls, approval gates, state updates, and subagent work behind durable boundaries, then test recovery and cancellation across the full call chain rather than relying on the agent SDK alone.

## Agent-Ready Context

Restate journals agent events so a failed run can resume without replaying completed work. Its demo combines a planner, human approval, parallel researchers, retries, cancellation, and session-isolated state; suspended approval waits consume **no serverless execution time**.

Treat production agents as persistent distributed processes. Put tool calls, approval gates, state updates, and subagent work behind durable boundaries, then test recovery and cancellation across the full call chain rather than relying on the agent SDK alone.

The evidence comes from a Restate demo, not a comparative production study. The cited **45 ms p99 for a 10-step workflow** describes its push architecture, but workload, deployment, and measurement details are not provided here.

## Connected Context

Feed7 judgment across 778 accumulated Signals:

This makes durable execution—not the agent SDK—the recovery boundary for production agents, concretizing prior calls for event logs and persistent sandboxes with resumable approvals, retries, parallel work, and cancellation. It strengthens the reliability pattern with a working serverless design, while the demo-only evidence and underspecified latency figure leave comparative performance and production generality unresolved.

- [Agent Frameworks Considered Harmful — Rémi Louf, .txt](https://feed7.dev/p/agent-frameworks-considered-harmful-remi-louf-txt-1ku9s0h) — Restate implements the candidate’s durable-event and replay requirements, while avoiding re-execution of work already recorded as complete.
- [What we’ve learned building cloud agents](https://feed7.dev/p/cloud-agent-lessons-06dh9iq) — Cursor’s Temporal rewrite provides independent production evidence that durable workflow infrastructure can materially improve cloud-agent reliability.
- [From fork() to Fleet: Designing an Agent Sandbox Cloud — Abhishek Bhardwaj, OpenAI](https://feed7.dev/p/from-fork-to-fleet-designing-an-agent-sandbox-cloud-abhishek-bhardwaj-op-0np9ki3) — Persistent sandboxes preserve execution state, while Restate preserves workflow progress; long-running agents may require both recovery layers.
- [Build with OpenAI Agents API on Vercel](https://feed7.dev/p/build-with-openai-agents-api-on-vercel-1xys7ei) — Both separate orchestration from isolated execution, and Restate sharpens the recovery, retry, approval, and cancellation behavior that must hold across that boundary.

## Context Map

- Layer: infra
- Domains: coding, research
- Topics: agent-reliability, observability, cloud-agents

## Uncertainty

- The evidence comes from a Restate demo, not a comparative production study. The cited **45 ms p99 for a 10-step workflow** describes its push architecture, but workload, deployment, and measurement details are not provided here.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.