Google AI Studio's Summer Sprint: Custom URLs, GitHub Import, and the Quiet Rebuild of Google's Dev Surface

Google AI Studio spent the last 30 days shipping small features that add up to a larger story: the platform is quietly becoming a full-stack app surface, not just a Gemini playground.

TLDR Between June 12 and July 10, 2026, Google DeepMind and Google AI Studio shipped at least seven distinct developer-facing changes: the Interactions API into GA, Design Variations for UI generation, chat-history sharing in Build, GitHub repo import, and — as of yesterday — free custom URLs under the ai.studio domain. Taken individually, each is a minor drop. Taken together, they reshape AI Studio from a model sandbox into a deployment target that competes with Vercel-style workflows for AI-native apps.

Key Takeaways

• Google AI Studio shipped free custom subdomains under ai.studio on July 10, 2026, removing Cloud Run as a prerequisite for shipping AI Studio apps.
• The Interactions API went GA on June 22, 2026 as Google's new default API for orchestrating across models and agents.
• GitHub import landed on July 8, 2026, closing the loop from external repo to deployed Build app.
• Design Variations, a one-click UI generation feature, launched June 26, 2026 and reportedly has themes support planned.
• The cluster suggests a coherent "vibe coding stack" strategy, but core specs — rate limits, model defaults for deployed apps, private-repo support — remain unverified.
• Google DeepMind's Gemini 3.1 Pro is listed as available in AI Studio, though which model powers deployed apps by default has not been stated.

What Actually Shipped in the Last 30 Days

The observable release cadence is the story. Working from the official Google AI Studio account and Logan Kilpatrick's rollout posts, here is what actually landed, in order:

June 12 — Information Agents in AI Mode Search go global. Robby Stein announced that Information Agents are available across all AI Mode languages and markets for Google AI Ultra subscribers. The framing is "your agent will work around the clock" — a persistent-background-agent surface, not a chat.

June 18 — AI Control Roadmap. Ashutosh Shrivastava surfaced DeepMind's AI Control Roadmap, framing advanced AI agents as insider-threat-adjacent and introducing a "dual controls" pattern — one AI acts, an independent system holds a brake. Positioning-heavy, but it sets Google's safety posture around agentic scale.

June 22 — Interactions API GA. Logan Kilpatrick confirmed that the Interactions API is now generally available and is "the new default API going forward." The stated purpose is to orchestrate across models and agents in a single interface.

June 25 — Gemma 4. Community praise for Gemma 4 surfaced, though no benchmark table accompanied the tweet. This is a community impression, not a Google claim.

June 26 — Design Variations. AI Studio rolled out Design Variations: a one-click button that generates several alternative UI layouts for a Build app. TestingCatalog reported that themes support is planned. {{EVIDENCE:2070629931080446364}}

June 29 — Chat-history sharing. Announced: when sharing an AI Studio Build app, developers can now include full chat history via a toggle, exposing the prompt workflow that produced the app. This is a small change with a large implication for prompt-transparency and reproducibility.

July 8 — GitHub Import. AI Studio enabled "Import from GitHub" — a one-click path from repo to Build. TestingCatalog flagged August 1st as a date to watch, apparently for mobile testing.

July 10 — Custom URLs. The most-viewed drop of the sprint. Apps deployed on AI Studio can now claim a personalized subdomain under ai.studio at no cost. The official post crossed 108,000 views within hours. Poonam Soni summarized the pitch bluntly: "clean, memorable, no Cloud Run required."

Why the Custom URL Drop Matters

A branded subdomain is a small feature with an outsized implication.

Before July 10, shipping an AI Studio app to a durable, presentable URL was not a solved problem. The typical path involved Cloud Run configuration, a Google Cloud project, and enough DevOps to scare off a designer prototyping with Design Variations. The Custom URL drop collapses that path to a click.

TestingCatalog's reaction — "I immediately remembered Google Websites from 2010. We went through a full cycle" — is the most useful frame available. Google has been here before with hosted app surfaces. The bet this time is that AI-native apps have a different distribution curve: they get made faster, forked more, and thrown away more casually than the static sites Google Websites hosted.

There is a related detail worth flagging. The placeholder URL your-own-url.ai.studio returns a 404, which the link snapshot confirms. This is expected — it's a template, not a live example — but it means Google has not yet published a formal URL syntax spec for the feature. Naming rules, reserved words, and takedown policy are all unstated.

The combined move — GitHub import on July 8, custom URLs on July 10 — is not a coincidence. Two days apart, the platform closed the loop from external repo to publicly reachable deployed AI app. That is a full-stack story compressed into a 48-hour window.

The Interactions API Is the Load-Bearing Piece

Buried in the sprint is a change most builders are underweighting: the Interactions API going GA.

Kilpatrick's framing — "our new default API going forward" and "the new era of Agents" — is the strongest positioning language in any of the observed announcements. The stated capability is orchestration across models and agents in a single interface. For anyone building multi-model pipelines today with hand-rolled routing between Gemini variants, Gemma, and third-party endpoints, this is the sanctioned pattern replacing that plumbing.

The Interactions API predates most of the sprint's surface-level features. Design Variations, GitHub import, and Custom URLs all sit on top of a Build layer that presumably calls into Interactions. A reasonable read: the API GA on June 22 unlocked the shipping cadence that followed.

Design Variations and the Vibe Coding Positioning

"AI Studio is becoming the one stop destination for vibe coding," Ashutosh Shrivastava wrote on July 10. The phrase captures the strategic posture of the sprint.

Vibe Coding, as the term is used in the community, means iterative app-building where the developer describes intent in natural language and lets the model handle scaffolding, styling, and glue code. Design Variations is a purpose-built Vibe Coding primitive: instead of writing "make it more modern and airy," a builder clicks a button and gets several concrete layouts to pick from. That is a substantive shift in what the interaction between developer and model looks like. {{EVIDENCE:2075595994327077095}}

The chat-history-in-share-toggle from June 29 fits the same posture. If the way you built the app is the app, then sharing the app must include the chat history. This is a Vibe Coding-native product decision, not a general software-engineering one.

Google DeepMind's Model Layer Beneath the Product Layer

The product surface is only half the story. According to the official Gemini 3.1 Pro model page, Gemini 3.1 Pro is currently in Preview status and available in Google AI Studio alongside the Gemini App, Gemini API, Gemini Enterprise Agent Platform, Google AI Mode, and Google Antigravity.

The published benchmark table on that page includes some sharp numbers. Gemini 3.1 Pro Thinking (High) posts 44.4% on Humanity's Last Exam with no tools, 77.1% on ARC-AGI-2 (ARC Prize Verified), 94.3% on GPQA Diamond, 80.6% on SWE-Bench Verified, and an Elo of 2887 on LiveCodeBench Pro. Its input context is listed at 1M tokens with a 64K output token limit and a January 2025 knowledge cutoff.

These are Google's own reported numbers on their own benchmark methodology, so they should be read as vendor claims until independently reproduced. The relevance to the AI Studio sprint: this is likely the model class powering the newly-deployable Build apps, though Google has not stated which Gemini variant is the default for deployed AI Studio apps.

Separately, DeepMind's FACTS Benchmark Suite — released December 9, 2025 with Kaggle — signals where Google's factuality evaluation is heading. The suite includes a Parametric Benchmark, a Search Benchmark, a Multimodal Benchmark, and Grounding Benchmark v2, totaling 3,513 public examples with a held-out private set. For anyone building on AI Studio and worrying about factuality regressions across model updates, FACTS is the leaderboard to watch.

Google AI Studio vs. Vercel v0: What the Signal Says

The most-mentioned comparison in the community isn't with another AI lab — it's with the class of Vercel v0-style AI app builders. Kilpatrick's own emphasis on "free apps, free deploys, and now free pretty URLs" reads as a direct positioning shot at pricing-forward competitors.

The signal set here is thin, so this is a scoped comparison against publicly stated AI Studio features only.

• Deployment cost. AI Studio: apps, deploys, and pretty URLs stated as free at rollout by Kilpatrick. Vercel v0 equivalent: unverified — no comparable public number from this signal set.
• Custom URL format. AI Studio: subdomain under ai.studio. Naming syntax not formally documented; the your-own-url.ai.studio placeholder returns 404. Vercel equivalent: unverified — no comparable public number from this signal set.
• Repo integration. AI Studio: "Import from GitHub" one-click. Private-repo, branch-selection, and push-back semantics not documented. Vercel equivalent: unverified — no comparable public number from this signal set.
• Model orchestration layer. AI Studio: Interactions API (GA June 22), Google's default going forward, orchestrates across models and agents. Vercel equivalent: unverified — no comparable feature described in this signal set.
• Design generation. AI Studio: Design Variations one-click UI generation, themes planned per TestingCatalog. Vercel equivalent: unverified — no comparable public number from this signal set.

The honest read: this signal set only supports comparing AI Studio's stated feature set against itself. The competitive frame is community-inferred, and any head-to-head numbers should come from a separate hands-on evaluation, not from vendor announcements.

What We Know vs. What We Don't

What we know

• Google AI Studio shipped free custom subdomains under the ai.studio domain on July 10, 2026, per the official announcement.
• Google AI Studio added "Import from GitHub" to Build on July 8, 2026, per the official post.
• The Interactions API went GA on June 22, 2026 and is Google's new default API for orchestrating across models and agents, per Logan Kilpatrick.
• Design Variations, a one-click UI-generation feature for Build apps, launched June 26, 2026 with themes support planned.
• Chat-history sharing was added to Build app sharing on June 29, 2026 via a toggle in the share flow.
• Information Agents in AI Mode Search rolled out to all languages and markets for AI Ultra subscribers on June 12, 2026.
• Gemini 3.1 Pro is listed as available in Google AI Studio on the official DeepMind model page in Preview status.

What we don't

• Which Gemini variant is the default for deployed AI Studio Build apps has not been specified.
• The formal URL naming syntax, reserved-word list, and takedown policy for ai.studio subdomains are not published.
• Whether GitHub import supports private repositories, branch selection, or push-back synchronization is not documented in the observed signals.
• Rate limits, quotas, and long-term free-tier terms for deployed apps and pretty URLs have not been disclosed.
• SLA guarantees or uptime commitments for ai.studio deployments have not appeared in any observed announcement.
• The August 1st date flagged by TestingCatalog is a community expectation, not a confirmed Google roadmap milestone.

Why This Matters for Builders

Three things change if this sprint is more than a coincidence of unrelated feature drops.

First, the marginal cost of shipping a prototype to a durable URL drops close to zero. That changes the calculus for hackathons, internal tools, and demo-driven sales cycles. If a builder can go from an idea to a shareable ai.studio URL in an afternoon, more prototypes will get shipped and the ratio of demos to production apps will shift.

Second, Google's model layer and product layer are now visibly co-designed. Interactions API GA on June 22 preceded and enabled the product-surface drops that followed. Builders picking an orchestration pattern today should factor in that Google is treating Interactions as the sanctioned surface, which affects the maintenance cost of hand-rolled routing.

Third, the Vibe Coding framing has moved from meme to product spec. Design Variations and chat-history-in-share are only sensible if you accept the premise that the interaction with the model is a meaningful part of the artifact. That is a design bet, and it is now shipping code.

What Builders Should Do Today

Three concrete moves.

Wire up a small test app in AI Studio Build and deploy it to a custom ai.studio URL. The point is to feel the shape of the deployment loop, not to ship anything durable. Look specifically for what happens when you update the app — does the URL re-deploy in place, or version?

Try the GitHub import path with a small existing repo and note what breaks. Does it handle a monorepo? What about lockfiles? Nothing in the announcement addresses these cases, so the answers will come from testing.

Read the Interactions API GA notes if you are currently orchestrating across models. Even if you do not migrate immediately, understanding what Google's default routing surface looks like will inform your next architecture decision.

The Week Ahead

Three signals to track.

Watch for an official Google AI Studio developer docs page describing the ai.studio custom URL format, including any naming or takedown rules. That page will convert the launch tweet into a durable spec.

Run a small deploy-and-share workflow yourself before making assumptions about rate limits or default model backing. Nothing in the observed signals answers those questions, and the vendor announcement is not a substitute for a live test.

Pin the August 1st date TestingCatalog flagged. Whatever Google is shipping around then — mobile testing surfaces for GitHub import is the current guess — is likely the next visible move in this sprint. If a separate model card or Gemini variant update lands the same week, the pattern of product-surface-plus-model-layer co-shipping will confirm.

Building similar AI-native app surfaces? On kie.ai you can try Gemini 3.1 Pro, Gemini 3.5 Flash, and Claude Opus 4.8.