---
title: "Discussion on \"I thought Session Replay recorded my browser agent. But it didn't\" | Hashnode"
description: "Discussion on \"I thought Session Replay recorded my browser agent. But it didn't\". An agent telling you it failed isn't the same thing as knowing where that agent failed.\n\nThat was the dilemma I faced..."
url: "https://hashnode.com/posts/i-thought-session-replay-recorded-my-browser-agent-but-it-didn-t/6ab16566d768452f5ac8fe86"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [I thought Session Replay recorded my browser agent. But it didn't](https://sunkanmi.hashnode.dev/i-thought-session-replay-recorded-my-browser-agent-but-it-didn-t?utm_source=hashnode&utm_medium=feed)

An agent telling you it failed isn't the same thing as knowing where that agent failed. That was the dilemma I faced...

Sep 21 · 11 min read · sunkanmi.hashnode.dev  
Read the article

-   [#ai-agents](https://hashnode.com/tag/ai-agents)
-   [#session-replay](https://hashnode.com/tag/session-replay)
-   [#posthog](https://hashnode.com/tag/posthog)

## Written by

[Sunkanmi Fafowora](https://hashnode.com/@SunkanmiFafowora)

@SunkanmiFafowora · Lagos, Nigeria

  
[View profile](https://hashnode.com/@SunkanmiFafowora)

Software Engineer / Technical Writer

Lover of all things tech. I write articles on Front-end development, JavaScript, React, Python, and anything that makes my life easier. Thanks for stopping by! 😊💛

## Responses 3

Markdown supported · ⌘↵ to post

[Rulestack](https://hashnode.com/@rulestack)

[Sep 22](https://hashnode.com/posts/i-thought-session-replay-recorded-my-browser-agent-but-it-didn-t/6ab16566d768452f5ac8fe86/comment/6ab2c24aaf33a75f59b93c16)

The misdiagnosis is the bit I recognise most: the recordings were there, filed somewhere the mental model didn't expect. We ended up with a blunt rule for that class of thing, a post counts as published only when the log line and the actual screen both show it, and either one alone is treated as a guess. It doesn't link the two the way you're after with the trace and the replay, it just refuses to trust one without the other, so I'm curious whether PostHog gives you any handle to stamp the session id onto the generation event yourself.

[Mihai\_LeanZero](https://hashnode.com/@mihaileanzero)

[Sep 21](https://hashnode.com/posts/i-thought-session-replay-recorded-my-browser-agent-but-it-didn-t/6ab16566d768452f5ac8fe86/comment/6ab18cc0a4f8532081d8da1d)

The Zod version detail is the one I'd want nailed down before trusting the fix. additionalProperties being required-and-false only matters once a provider's tool-schema strict mode enforces it, and Zod v4 changed how object schemas serialize compared to v3, so if Stagehand's schema generation assumed v3 semantics, pinning back to 3.25.76 is believable as the real fix rather than a side effect of the restart. Worth checking whether Stagehand documents a supported Zod range anywhere, because "downgrade one patch version and restart" is exactly the kind of fix that quietly breaks again on the next npm install. The Activity-vs-Session-Replay split deserves more than a footnote too, it's a real gotcha for anyone pointing an agent at PostHog and expecting the human-facing dashboard to just work.

[Sunkanmi Fafowora](https://hashnode.com/@SunkanmiFafowora)

[Sep 25](https://hashnode.com/posts/i-thought-session-replay-recorded-my-browser-agent-but-it-didn-t/6ab16566d768452f5ac8fe86/comment/6ab18cc0a4f8532081d8da1d#reply-6ab68b888d50d5d8ace629c2)

Hmmmm. Stagehand's internal schema gave that error.

It was rejected by OpenAI's structured and strict output format. The thing is, even though I pinned Stagehand's version to `v3.25.76`, `v3.7.3` works and supports both Zod 3 and 4.

See, the reinstall did change the dependency graph, but the pinned graph still contains the Zod 4 requirement; it gets confusing at a point as to why this is the case, but I'll treat my solution as a workaround rather than a confirmed long-term fix.

I also saw a [Stagehand](https://github.com/browserbase/stagehand/issues/1845) issue showing that Zod 3/4 can still break a schema conversion in Mastra builds.

All this seems like a solution is still being worked on, this is just a temporary workaround.
