# Callee Interfaces Outlast Caller Integrations Under Provider Churn ## Claim A platform that exposes itself as a standardized, permission-gated callee keeps its AI-agent compatibility even as the underlying model providers change, while a platform that only calls out to a specific AI provider has to be re-integrated every time that provider relationship shifts. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds for any software platform choosing between "add an AI feature" (caller architecture, calling out to a chosen model API) and "expose a stable, machine-operable interface" (callee architecture, letting any agent call in against a typed, permission-gated surface). - Breaks down if the standard itself (here, MCP) fails to achieve durable multi-vendor adoption — the claim assumes the callee interface is built on a standard with staying power, not a proprietary one-off. - Doesn't establish which specific implementation (WordPress's Abilities API/MCP Adapter or any other) executes this well, only that the architectural choice matters independent of implementation quality. ## Evidence WordPress 7.0 shipped an AI Client (a provider-agnostic PHP API for connecting to Anthropic/Google/OpenAI models — the caller role) alongside a matured Abilities API and an official MCP Adapter plugin (the callee role: a typed, permission-gated registry of site capabilities that any MCP-speaking agent can call into). Developer Jorijn Schrijvershof's caller/callee distinction, quoted in MemberPress's market-share analysis, frames this dual positioning as the strategically decisive move: a site owner's AI-feature investment survives the next model-provider shakeout because the site's capabilities are exposed once, standardly, rather than wired to whichever provider built the feature. ## Links - [[3. Resources/Permanent Notes/Caller Distance Determines the Right WordPress Interface]] — because contrasts: that note picks among WP-CLI/REST/Abilities API by caller proximity to the site; this note argues the callee-vs-caller architectural choice itself is what survives provider churn, a level up from interface selection. - [[3. Resources/Permanent Notes/A Network Is Open Only When Its Parts Are Independently Replaceable]] — because extends: callee-interface durability is a specific case of the general principle that independently replaceable parts outlast tightly-coupled ones. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in: platform strategy for surviving rapid AI-provider turnover. - [[3. Resources/MOCs/WordPress Platform Design MOC]] — applies-in: the MOC lists this note under Synthesis additions. ## Tests / falsifiers - If MCP itself fails to reach durable multi-vendor adoption in the next 12–24 months (per the article's own hedge), platforms that bet on it as their callee standard lose the advantage this claim predicts — worth revisiting on that timeline. - If a caller-only platform proves just as easy to re-point at a new provider (e.g., via a thin abstraction layer) as a callee platform is to keep current, the durability gap this claim asserts doesn't hold. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Literature Notes/WordPress Lost Market Share (MemberPress)]] — reports the caller/callee framing and WordPress 7.0's AI Client/Abilities API/MCP Adapter shipment this claim is drawn from.