---
title: "Discussion on \"Self-healing tests are table stakes, and most of them aren't real\" | Hashnode"
description: "Discussion on \"Self-healing tests are table stakes, and most of them aren't real\". Every vendor claims self-healing tests. Most mean selector retries. Here's the mechanism that separates real healing from a deferred break, and its limits."
url: "https://hashnode.com/posts/self-healing-tests-real-vs-fake-2026/6a4ff32072b571606b890d4c"
---

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

Discussion

# [Self-healing tests are table stakes, and most of them aren't real](https://bug0.com/blog/self-healing-tests-real-vs-fake-2026?utm_source=hashnode&utm_medium=feed)

Every vendor claims self-healing tests. Most mean selector retries. Here's the mechanism that separates real healing from a deferred break, and its limits.

Jul 9 · 7 min read · bug0.com  
Read the article

-   [#self-healing-tests](https://hashnode.com/tag/self-healing-tests)
-   [#test-automation](https://hashnode.com/tag/test-automation)
-   [#flaky-tests](https://hashnode.com/tag/flaky-tests)
-   [#ai-testing](https://hashnode.com/tag/ai-testing)
-   [#qa-automation](https://hashnode.com/tag/qa-automation)

## Written by

[Syed Fazle Rahman](https://hashnode.com/@fazlerocks)

@fazlerocks · San Francisco, USA

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

Co-founder, Hashnode and currently building Buildful and Bug0.

Co-founder and CEO at Hashnode

## Responses 3

Markdown supported · ⌘↵ to post

[Farhana Yasmin](https://hashnode.com/@Farha)

[Jul 10](https://hashnode.com/posts/self-healing-tests-real-vs-fake-2026/6a4ff32072b571606b890d4c/comment/6a509bc0b8dd8961c2cafd0c)

Good taxonomy, but I'd argue intent re-resolution doesn't eliminate brittleness, it relocates it. A selector is brittle against the DOM. "Add the item to the cart" is brittle against ambiguity. On a page with a quick-add button, a wishlist icon, and a buy-now CTA, the model is making a judgment call every time it re-resolves. The dual-model assertion helps, but two models agreeing is still probabilistic, not deterministic. Curious how you handle steps where the intent itself is ambiguous.

[Jake Morrison](https://hashnode.com/@jakebuilds)

[Jul 10](https://hashnode.com/posts/self-healing-tests-real-vs-fake-2026/6a4ff32072b571606b890d4c/comment/6a509c005ed2123625e888fa)

The false green section hits. We had a retry-based setup mask a broken coupon flow for 11 days because the test kept passing on the second attempt. Nobody looks at a green suite. Support tickets found the bug before CI did. Since then I trust a loud failure over a quiet pass every time.

[Marcus Chen](https://hashnode.com/@dev_marcus)

[Jul 10](https://hashnode.com/posts/self-healing-tests-real-vs-fake-2026/6a4ff32072b571606b890d4c/comment/6a509be92f00cc424fde9d94)

The vendor question at the end is the useful bit. I asked our current tool's support team almost exactly this last quarter and got a healing percentage back, which tells you something. One thing the post doesn't cover: what does a redesign actually cost under intent re-resolution? If 200 cached steps invalidate at once, that's 200 AI resolutions in one run. Is that minutes or hours, and who pays for the tokens?