We expect Stitch Loop's waitlist or Google AI Ultra priority to remain in place on 30 June 2027; a 100 PR scenario prices rejected reviews at $1,495 a month.
A checkout conversion metric slips after a team sets up a loop to monitor it. Google's new Stitch Loop is designed to notice, work on a fix in a cloud sandbox and send a GitHub pull request to an engineer. The metric then becomes an invitation to make a decision, with a code change attached. If the metric slips again, the invitation returns. Having software watch a product continuously is different from having an engineer available to approve everything it proposes.
We would budget for Stitch Loop remaining behind a waitlist or Google AI Ultra priority on 30 June 2027. Our explanation is Split Costs: Google runs the agent in its cloud sandbox while customers' engineers review its pull requests. Each new user expands Google's standing workload and, potentially, somebody else's review queue. We expect controlled access to let Google study both loads before broad distribution. It also gives customers time to decide whether the proposed fixes earn more than their approval process costs.
In its 18 November 2026 announcement, Google describes a product agent that connects GitHub to logs, user behavior and feedback using more than 180 integrations, custom MCP servers and local CLI uploads. Three default loops look for cybersecurity, latency and accessibility problems; customers can create others, including checkout conversion. The agent proposes suggestions and solutions, builds code in an isolated cloud sandbox, and reflects on the result. It also shares persistent product context with coding agents such as Google Antigravity through the Stitch CLI and MCP, and supports Slack digests and new loops. When a goal is met, it continues watching for a reversal. GitHub pull requests can include screenshots and video, but Google's rule is: "Nothing merges until you review the PR." The machine is persistent; the authority to ship remains human.
Google opened a global waitlist on launch day and gives eligible Google AI Ultra users priority. On the US subscription page read 18 November 2026, Ultra starts at $99.99 a month for five times Google AI Pro's usage limits, while $199.99 buys twenty times those limits. Those are Ultra subscription prices, and Google has announced no separate Stitch Loop price; they give little guidance about the engineer's time after the agent starts producing work.
Reviewers encounter their own costs well before any proposed fix reaches production. A study published 11 June 2026 of agent written fixes found 1,497 rejections among 3,225 pull requests from Copilot, Devin, Cursor and Claude, or 46.41%. Incorrect implementation, failing continuous integration and low priority were among the reasons. Take a scenario of 100 fix requests a month, with the same rejection rate and an assumed 30 minutes to review each rejection. One hundred times 46.41% times 30 divided by 60 comes to 23.2 reviewer hours; at the 2025 US Bureau of Labor Statistics median rate of $64.44 an hour, the bill is $1,495 per month. It is about 15 times the $99.99 Ultra entry price, which makes reviewers a larger budget item than the subscription in this scenario.
Those pull requests came from other agents working on open source repositories, rather than from Stitch Loop against live products. The scenario prices decisions about rejected work, rather than claiming every rejection is worthless. A separate study found Cursor's fix requests accepted at 80.4%, and documentation requests at 82.1%, against 66.1% for new features. At 80.4% acceptance, 100 requests leave 19.6 rejected; at 30 minutes each, 9.8 hours times $64.44 costs $631 a month. Even the better selection rate leaves a substantial recurring bill. At 100 requests and the 46.41% rejection rate, the $99.99 Ultra entry price equals 2.0 minutes of review per rejected request ($99.99 times 60 divided by (46.41 times $64.44)).
Jules is what makes our June 2027 prediction risky. Google introduced Jules in Labs in December 2024 and declared it "available to everyone. No waitlist." on 20 May 2025, about five months later. By 6 August 2025, Jules had left beta after more than 140,000 code improvements shared publicly, with Pro and Ultra usage tiers. Google's precedent is a rapid expansion with paid limits still attached. Alphabet's quarterly filing shows $20,640 million of research and development spending on $127,577 million in revenue for the quarter ended 30 September 2026. Google has money to fund a launch before the business is optimized.
A better agent can reduce the queue before it reaches a senior engineer. Google DeepMind said on 6 October 2025 that CodeMender, now paired with Gemini 4 Carbon in Loop's security work, upstreamed 72 security fixes during six months of development and surfaced high quality patches for human review. TestingCatalog, after testing Loop for weeks, called the experience "a higher level of delegation" and observed changes to its site's theme and suggestions about accessibility and compliance. Combined with Cursor's 80.4% fix acceptance, this is the attractive version of the opposing case: select useful work early, save engineer time and open access quickly.
Yet Jules and Loop ask Google to distribute different workloads. A Jules user starts a bounded coding task; Google's Loop pitch is a process that monitors production metrics 24/7 and returns to a problem after an apparent improvement slips. More than 180 possible integrations add context, along with continuing work in sandboxes and human approvals. The comparison that matters through 30 June 2027 is five months to open a task agent versus the decision to open a persistent service to everyone. At 100 proposed fixes monthly, even the 80.4% acceptance scenario still uses $631 of reviewer labor. An unrestricted signup page alone will not create more engineering hours for its customers.
Google says millions of people already use Stitch to build interfaces, giving Loop an established audience. Split Costs also changes which customers look attractive. A heavy user who finds many valuable fixes can be the customer with the most ongoing agent work. A small team may value the product but struggle to give each proposed change a proper hearing. Google therefore has a reason to control exposure while it improves selection and decides how usage should be priced or limited. The distinctive product advantage would be deciding which promising problems justify an engineer's attention before generating another approval request. That discipline makes access more valuable than sheer volume of suggestions.
We would plan for restricted access and treat engineering review as part of the purchase decision, rather than price the service from the Ultra subscription alone. We would be wrong if a dated Google Labs announcement or the Loop access page shows on 30 June 2027 that Loop is available to everyone without a waitlist and without Ultra priority.
