# Google opens Stitch Loop waitlist for a product agent
> The experiment connects live product signals, proposed changes and human review.
By BIGCHANGE.AI Editorial
Published: 2026-11-20T00:15:37.953Z
Updated: 2026-11-20T00:15:37.953Z
Canonical: https://bigchange.ai/blog/google-stitch-loop-product-iteration

AI-generated conceptual illustration for BIG CHANGE.
Google Labs announced Stitch Loop on November 18, an experiment connecting a codebase with product logs, usage and feedback to propose improvements after launch. Its waitlist is open globally, with eligible Google AI Ultra users given priority. [Google announcement](https://blog.google/innovation-and-ai/models-and-research/google-labs/stitch-loop/)
## The big change
- Google describes an agent that stays with a shipped product, using operational evidence to propose changes and observe results. Its job is intended to run from identifying a useful change through checking the result.
- A pull request gives a person a specific change to examine. Reviewers can discuss the implementation, approve it or request revisions, while repository rules determine which approvals are required before merging.
- Judging an improvement requires a defined measure and evidence about cause. A comparison group can help a team evaluate whether a release changed customer behaviour, informing which idea it should pursue next.
## From a product signal to a reviewed change
Google describes four stages. Loop surfaces suggestions from product evidence, then offers solutions with scope and trade-offs. Once a user chooses a direction, it writes code in an isolated cloud workspace and opens a GitHub pull request. Google says the user reviews the request before anything merges. Loop then monitors live data, uses positive and negative results to inform later suggestions, and continues watching after a goal is met. Checkout conversion is one of Google's examples of a custom loop. [Google announcement](https://blog.google/innovation-and-ai/models-and-research/google-labs/stitch-loop/)
## Which shoppers enter the comparison?
Consider a hypothetical shop where many people start checkout but relatively few complete a purchase. The team might investigate shipping charges or a broken button, two possible explanations in Google's Analytics guidance. A proposed button repair is a hypothesis about what could help customers finish. [Purchase journey documentation](https://support.google.com/analytics/answer/13128171)
Google Analytics' Purchase journey report traces users through events including product views, adding to a cart, beginning checkout and purchasing. Its default closed funnel counts users who follow each step in the specified sequence. An open funnel allows entry at later stages. Switching between them changes who appears in the later counts. [Purchase journey documentation](https://support.google.com/analytics/answer/13128171)
For the hypothetical shop, a completion measure therefore needs a consistent group of checkout starters and a matching count of purchases. Keeping the same event definitions and funnel conditions makes the comparison interpretable when assessing a later release. [Purchase journey documentation](https://support.google.com/analytics/answer/13128171)
## Review applies to the actual edit
GitHub's review documentation describes three decisions, comment, approve and request changes. A reviewer can discuss specific lines and suggest a correction. For a proposed checkout repair, that gives a person a place to examine what the code changes and ask the author to address a concern. [GitHub review documentation](https://docs.github.com/en/pull-requests/reference/pull-request-reviews)
GitHub also documents configurable rules requiring a specified number of approvals. Repository administrators can require an approval after the most recent reviewable push or dismiss an earlier approval when the changes differ. That connects review to the current edit as more code arrives during discussion. These are repository controls a team can configure around its development process. [GitHub rules documentation](https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/managing-rulesets/available-rules-for-rulesets)
## The metric's movement still needs an explanation
Microsoft researchers' *Online Experimentation at Microsoft* describes randomly assigning users to a control or changed version. With an appropriately designed experiment, the groups provide a way to test whether the change caused a difference in behaviour. The paper explains why effects outside a release can overwhelm a comparison made only across successive periods. [Original research](https://www.microsoft.com/en-us/research/publication/online-experimentation-at-microsoft/)
For a team deciding what to build next, the useful question is how the changed experience performed against a comparable unchanged experience. The experiment's design and statistical evidence determine the conclusion. [Original research](https://www.microsoft.com/en-us/research/publication/online-experimentation-at-microsoft/)
BIG CHANGE reviewed the announcement and documentation. We have not tested Loop.
## Sources
- [Google Stitch Loop announcement](https://blog.google/innovation-and-ai/models-and-research/google-labs/stitch-loop/) — November 18 workflow and waitlist.
- [Google Analytics purchase journey](https://support.google.com/analytics/answer/13128171) — Funnel steps and population definitions.
- [GitHub pull request reviews](https://docs.github.com/en/pull-requests/reference/pull-request-reviews) — Review decisions and discussion.
- [GitHub repository rules](https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/managing-rulesets/available-rules-for-rulesets) — Configurable approvals for current changes.
- [Online Experimentation at Microsoft](https://www.microsoft.com/en-us/research/publication/online-experimentation-at-microsoft/) — Original 2009 research on controlled comparisons.
The BIG CHANGE newsletter
The big picture. At your pace.
Recent stories on AI and robotics, the shifts worth watching and practical ideas to use. Choose a daily briefing, weekly digest or monthly perspective.
Next scheduled send (UTC): . Your first edition arrives at the next scheduled send after you confirm.
Your privacy, your choice.
Necessary storage supports site security and remembers your choices. Optional Google Analytics stays off until you allow it. You can read every story with necessary storage only. Privacy details