Apple Is Quietly Engineering an AI Agent Carve-Out for the App Store. Here's the 4-Week Prep Window Before WWDC.

On May 13, Mark Gurman reported in Bloomberg that Apple is working on a way to allow autonomous AI agent apps into the App Store while keeping the existing review process and privacy guardrails intact. Apple's current review guidelines effectively ban what Gurman calls "vibe coding tools" and most agentic AI apps that generate behavior on the fly. WWDC is June 8–12, with the keynote on Monday, June 8.

The reporting is "exploratory" — no confirmed timeline, no confirmed feature set. The natural reveal window is WWDC, but Gurman explicitly noted there is no guarantee anything ships at the keynote. Apple has a history of announcing policy work and then taking 12 to 18 months to land the actual change.

That uncertainty is exactly why the prep work matters now and not in three weeks. If a carve-out does land, the indie devs in the first 30-day submission window get the same first-mover advantage that Instagram and Tapbots got out of the original 2008 App Store launch. The optionality is cheap to preserve right now and expensive to manufacture after the fact.

What Bloomberg actually reported

Apple engineers are designing a system that would allow AI agent apps to operate inside App Store rules while preserving privacy, security, and behavioral review. The current explicit ban is on tools that let developers create dynamic mini-apps on the fly — the "vibe coding" category — and any app whose behavior is not fully specified at submission time.

The carve-out, if it ships, would create a new review category. The expected shape: stricter sandboxing of agent capabilities, mandatory disclosure of what the agent can do and what data it accesses, possibly a separate review track with different timing and different reviewer guidelines. None of that is confirmed. All of it is the obvious set of guardrails that would let Apple thread the needle between "agents are interesting" and "agents can't be allowed to break our existing product safety model."

The forcing function is that third-party AI agent apps are already in the App Store via the back door. ChatGPT, Claude, Perplexity all ship as conversational chat apps and have progressively added agent capabilities — calendar access, web browsing, file manipulation, on-device tool use. Apple's choices are now narrowed to three: ban the back door and lose those apps to the web and Android, ignore it and watch the agent surface area expand without explicit policy, or build a proper review framework. The carve-out is the third option materializing.

The four-week prep that's worth doing this weekend

The work below is useful whether the carve-out lands at WWDC, ships later, or never happens at all.

If you have been sitting on an iOS AI agent idea, the right time to spec the product is right now. Not the build — the spec. Write down what the agent does, what it accesses on the device, what it stores off-device, what user permissions it requires, and what the disclosure story to the user looks like. If the carve-out drops at WWDC with the guardrails I'd expect, this disclosure document becomes the core of your submission paperwork. If the carve-out doesn't drop, you still have a tight spec for the next time the opportunity opens.

App Store review backlogs at a new category launch are weeks-to-months. The first wave of accepted apps is the wave that submitted in the first 30 days, because the reviewer pool is sized for steady-state volume and gets overwhelmed by the launch surge. If you have a TestFlight build of an agent app at the keynote, you submit on day one. If you start writing the app at the keynote, you submit in week six, in a queue that's three months deep.

The keystone preparation is having a working build of the boring half — auth, persistence, basic UI, App Store metadata, privacy nutrition label — before the policy clarifies. Then the agent logic is what you add in the four-week sprint after the announcement. Without the boring half done, the four weeks aren't enough.

Audit your existing apps if you ship anything in the store today

If you already have apps in the App Store, this is the underrated prep work. Apple's category transitions historically don't grandfather behaviors that the new policy explicitly addresses.

Look at every behavior in your existing apps that could be characterized as agent-adjacent. Autonomous task execution. Background polling that takes action. Agent-style instruction following from natural language input. The cases where your app interprets user intent and takes a multi-step action.

For each one, ask: under a strict new agent-app review, would this behavior require disclosure I haven't done, or pass review under the existing category I'm in? Document the answer. If a new policy lands and your existing app gets re-categorized, you want to be the developer who already knows what changes, not the developer who finds out from a rejection email.

The specific guardrails I'd expect — Sign in with Apple where any account is involved, App Tracking Transparency for any cross-app data, explicit consent flows for tool use that touches user data — are the ones that grandfathered violations rarely survive. If your existing apps cut corners on any of these, this is the window to clean it up.

The distribution math, plainly

The App Store has approximately 1.5 billion active iOS users globally. iOS users spend roughly two to three times more per app on average than Android users do. For an AI-agent-shaped product, the App Store represents the largest single distribution surface that does not currently accept your category at all.

If Apple actually ships the carve-out — and shipping it at all is the question, separate from when — the indie devs with submission-ready apps in the first 30-day window get a structural advantage that closes quickly. The big-tech agent apps (ChatGPT, Claude, the eventual Apple Intelligence agent apps) will dominate the category within 90 days. The window for indie apps to establish category leadership is narrow and front-loaded.

That math is the reason the prep work pays off so unevenly across timing. A weekend of work now positions you for a 12-month distribution advantage if the carve-out lands as reported. A weekend of work after the keynote positions you for a fifth-place finish behind the apps that were already submission-ready.

The honest counter-take

Gurman's reporting is reliable but not infallible. Apple's pace on App Store policy changes is typically slow. "They're exploring a thing" has a long history of meaning "they're exploring a thing and it ships 18 months later, narrower than the original report suggested."

The recent track record on App Store category expansions is not encouraging. EU sideloading shipped narrower than the original announcement. Vision Pro's visionOS app store launched with fewer apps than anyone expected and the category has struggled to grow. Web distribution for iOS apps in the EU is technically real and effectively unused. The base rate for Apple announcing a category change at WWDC and shipping it as announced is lower than the press cycle implies.

The reasonable prior is that if the carve-out lands at WWDC, it ships with restrictions that disappoint most of the apps that wanted to ship under it. The right framing of the four-week prep isn't "build for the WWDC announcement." It is build for the optionality of an iOS distribution channel that becomes a 2-week sprint when the policy clarifies, rather than a 6-month rebuild. The optionality is cheap. The certainty is not coming.

What I'd actually do this week

If you have an iOS-shaped AI agent product in mind, write the spec. Privacy disclosures, capability list, permission model. One document, two pages. This is the cheapest part of being ready.

If you have apps already in the store, run the audit. Behavior-by-behavior, look for anything that would re-categorize under an agent-app review. Make the list. Don't fix it yet — fix it after the policy lands and you know what specifically to fix.

If you don't have an agent product in mind and weren't planning one, skip this entire post. The App Store distribution surface is real and meaningful but the prep work only pays off if you have something to put on it. Building speculatively for a possible category that ships in a narrower form than reported is the worst use of a developer weekend.

The right prep is not betting the company on the carve-out shipping as Gurman described it. The right prep is making sure that if it does ship, you are not the developer who learned about the launch from someone else's launch post.

Sources

• Apple Working on Plan to Allow AI Agent Apps on the App Store — MacRumors
• Apple Prepares App Store for Autonomous AI Agents — PYMNTS
• Apple is working to incorporate AI agents on the App Store, per report — 9to5Mac
• Apple Is Figuring Out How to Let Agentic AI Into the App Store Without Breaking Everything — Android Headlines
• Apple may be building an AI App Store — and it could change the iPhone forever — Tom's Guide

Fact-check log

• Mark Gurman / Bloomberg reporting May 13 → verified (MacRumors, 9to5Mac, PYMNTS all credit Bloomberg/Gurman)
• Apple's current ban on "vibe coding tools" → verified (Gurman's reporting, multiple secondary sources)
• WWDC 2026 dates → corrected from original "June 9–13" to "June 8–12" (Apple Newsroom: WWDC returns week of June 8, keynote Monday June 8, 2026, confirmed by Apple Developer site)
• "Exploratory" status, no confirmed timeline → verified (Gurman's framing in all sources)
• ~1.5 billion active iOS users → reasonable; Apple's 2025 reports cited ~2.35B active devices total across all product lines, iPhone active devices roughly 1.4–1.5B; framed as "approximately"
• iOS users spending 2–3x more per app vs. Android → consistent with industry reports (Sensor Tower, data.ai); softened wording slightly
• ChatGPT, Claude, Perplexity already in App Store as chat apps with agent capabilities → verified
• Recent App Store category expansions shipping narrower than announced (EU sideloading, visionOS, web distribution) → verified by industry coverage
• 30-day first-wave submission advantage at category launches → analytical claim from App Store launch patterns, framed as analysis Run: 2026-05-15

Voice-check log

• "uneven leverage" (line 56) → rewritten to "pays off so unevenly across timing" to avoid the LLM-tell "leverage"
• All H2 headings in sentence case → verified
• LLM-tell scan (delve into / unlock / seamless / cutting-edge / robust / revolutionize / game-changing) → no remaining hits
• First-person presence → "I'd expect" multiple times in the policy-analysis sections; "What I'd actually do this week" section closes the post
• Honest-take section present → "## The honest counter-take" with explicit track-record critique of recent Apple category launches (EU sideloading, visionOS, web distribution)
• Concrete recommendation → spec the agent product now, audit existing apps for agent-adjacent behaviors, prepare submission paperwork for first 30-day window
• Em-dash density → checked, varied rhythm
• Three-item power lists → none
• Summary conclusion → no — closes on a specific "right prep" framing
• "It's important to note" / "It's worth mentioning" → none Run: 2026-05-15