Source: https://www.thetechstack.com/moves/2026-09-15-agent-recovery Author: Kyle Brierley Author profile: https://www.thetechstack.com/authors/kyle-brierley Editor: Not specified Editor profile: Not specified Publisher: TheTechStack (https://www.thetechstack.com) Published: 2026-09-15 Updated: Not specified Suggested citation: [Kyle Brierley](https://www.thetechstack.com/authors/kyle-brierley). [Before choosing an agent platform, test what survives a failure](https://www.thetechstack.com/moves/2026-09-15-agent-recovery). TheTechStack. Retain the source qualifications. Cite the original article and link the credited author profile where available. An editor is not necessarily the author. **Stack move**Published 2026-09-15 # Before choosing an agent platform, test what survives a failure A research brief on session recovery, isolated execution, and the questions to bring to an agent-platform evaluation. By [Kyle Brierley](https://www.thetechstack.com/authors/kyle-brierley) Record Published 2026-09-15 Format Source-backed research brief Source and proposed tests are labeled in the brief. This is not an independent product test. **Research brief · Source published April 8, 2026 · Checked September 15, 2026.** This is an analysis of an existing engineering publication, not a new product announcement or an independent platform test. ## The documented architecture Anthropic describes separating the agent loop, execution environment, and durable session history in Managed Agents. Its engineering account explains how a replacement agent loop can recover history after a crash, and how separating execution from credentials changes the security boundary. These are vendor-described design choices, not results independently reproduced by TheTechStack. [Read Anthropic’s engineering account](https://www.anthropic.com/engineering/managed-agents). ## Our take: evaluate the recovery path Teams comparing agent platforms should examine failure behavior alongside successful demos. Ask the supplier to demonstrate recovery with a harmless test workflow. Record the platform version, configuration, and observations rather than accepting a general assurance that work resumes. Use these questions as an evaluation worksheet: - **Execution stops:** Which state persists, and what must be recreated? - **An action succeeds but its response is lost:** How does recovery avoid performing the action twice? - **Access is revoked mid-task:** Can a resumed task still reach the resource? - **A reviewer approves a draft that later changes:** Does the earlier approval remain valid? - **The operator investigates an incident:** Can they reconstruct the steps without exposing secrets? These are our proposed evaluation questions. The linked article does not establish that any product satisfies all of them. ## Who should use this Product and engineering leads evaluating agents that run across multiple steps or touch business systems. For a simple drafting assistant, start with a smaller test plan focused on answer quality and review. For an agent that changes records, document recovery and duplicate-action behavior before expanding access. ## Evidence still needed A platform architecture description does not establish fit for your workflow. Tenant-specific permissions, supported integrations, retention and export behavior, operational cost, and failure handling need current documentation or a recorded demonstration. No deployment recommendation or measured performance claim is made here. [Map these responsibilities in the AI stack guide](https://www.thetechstack.com/guides/ai-stack), or [capture the unresolved decisions in your pilot review](https://www.thetechstack.com/workbench). [Kyle Brierley](https://www.thetechstack.com/authors/kyle-brierley) · Founder, TheTechStack Research and drafting use AI assistance. Kyle is responsible for editorial direction and corrections. Source-backed observations, recommendations, and proposed tests are identified separately. [About Kyle and more of his research →](https://www.thetechstack.com/authors/kyle-brierley) ## Put the research to work Use the documented patterns as questions for your own team. Your systems, permissions, and acceptance criteria still need to be checked. - [Compare build, buy, and extend](https://www.thetechstack.com/guides/build-buy-or-extend)Work through capability fit, integration effort, cost, and ownership. - [Find a workflow like yours](https://www.thetechstack.com/use-cases)Explore knowledge search, support drafting, and sales RFP responses. - [Create a pilot review](https://www.thetechstack.com/workbench)Leave with a shareable brief and a list of open decisions. - [Evaluate a pilot](https://www.thetechstack.com/guides/evaluate-an-ai-pilot)Compare actual results with the current process, including review and rework. ## Follow the research Get new research briefs and company teardowns by email, in your feed reader, or on X. [Subscribe by email →](https://buttondown.com/thetechstack)[Subscribe via RSS →](https://www.thetechstack.com/rss.xml)[Follow on X →](https://x.com/thetechstackHQ) Email signup is hosted by Buttondown. RSS and X are available without an email subscription.