A year after launching apps inside ChatGPT, OpenAI folded them into a Plugin Directory and paired them with scheduled tasks — the makings of repeatable, in-chat workflows. Here's what's real for creators, and where it stops.
2026-09-29 · by Moe Ameen
The first half of this story landed at OpenAI's DevDay on October 6, 2025, when the company launched apps inside ChatGPT alongside a preview of the Apps SDK. Instead of returning plain text, ChatGPT can render a third-party tool as an interactive interface directly in the conversation — an inline card, a full-screen view, or a picture-in-picture panel — and you invoke it by name (e.g. "Spotify, make a playlist for my Friday party") or ChatGPT surfaces a relevant app in context. The apps are built on the Model Context Protocol (MCP), the open standard OpenAI adopted for connecting models to outside data and actions. Launch partners included Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow.
The platform matured over the following year. OpenAI opened third-party app submissions on December 17, 2025, with a reviewed App Directory at chatgpt.com/apps and approved apps rolling out to users through early 2026; the directory grew to well over a thousand apps. Then on July 9, 2026, OpenAI consolidated the App Directory into a new Plugin Directory. A plugin is the package: it can bundle one or more apps together with "skills" and app templates aimed at a specific workflow, so discovery shifted from single apps to workflow bundles. Existing connections kept working with nothing to resubmit, and the directory is available across ChatGPT on web and desktop, in ChatGPT Work, and in Codex. The "plugins" label echoes ChatGPT's 2023 plugin era, but this is the newer, MCP-based system, not a revival of the old one.
The other half is automations. ChatGPT's scheduled tasks, introduced in beta in early 2025, let you set a prompt to run automatically — once at a set time, or on a recurring daily, weekly, or custom cadence — and come back with the result. Through 2026 OpenAI expanded them with web monitoring (alert me when a page changes) and integrations with supported apps such as Gmail, Slack, and GitHub, and opened scheduled tasks to free users with a small number of active tasks that scales by plan. In practice it is a reminder-and-monitor tool: it is roughly capped to about once an hour, and a task can't use voice, files, or Custom GPTs. It mostly tells you things or nudges you rather than executing a complex job end to end.
Put together, apps and automations are the two halves people mean by "repeatable content workflows in ChatGPT": an app gives you an interactive tool inside the chat, and an automation reruns a prompt on a schedule. The honest read for a creator is to separate what this acts on and displays from what it makes. An app renders a partner's interface and can trigger that partner's actions; an automation re-runs your instructions on a timer. Neither renders branded video, images, or carousels, holds a brand voice you configure once, sizes a post per platform, or publishes to your social feeds. Treat specific partners, limits, and dates as a snapshot and confirm details on OpenAI's own pages.
Here is the gap to write into. OpenAI just made it easier than ever to build a repeatable workflow inside ChatGPT — an app you invoke by name, a task that fires every morning — and creators are asking whether that finally runs their content. It doesn't, and the reason is clean: an app displays a partner's interface and an automation re-runs a prompt, but the day's actual deliverable is finished, on-brand media sized per platform and posted to your feeds, and none of that lives inside the chat. That distinction is the whole story, and it's the one your audience wants explained.
So cover the news with the tool that closes the gap. Take your angle into [Kompozy](/) and one take becomes a week of posts: a native Text Post for X and LinkedIn on "what's real vs. what still lands on you," a [Carousel](/glossary/hyperframes) mapping the apps half against the automations half, a [Quote Graphic](/glossary/output-buckets) pulling the "reminder, not a publisher" line, a captioned [Persona Short](/glossary/persona-shorts) explaining the difference to camera, a Blog Article for the long-tail search, and a newsletter — all held to one voice by a [Persona Brief](/glossary/persona-brief) and shipped across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot). The sharper point converts because it's true: a ChatGPT automation is a scheduled prompt; Kompozy's Autopilot is a scheduled *publish* — it generates the media, holds the brand, and posts it, on a cadence, without you in the loop. If you want to keep drafting or brainstorming inside a ChatGPT app, do — then hand the result to Kompozy to render and distribute. For the evergreen breakdown see [ChatGPT Apps & Automations](/ai-tools/chatgpt-apps-and-automations); if you're deciding whether to build your content workflow in ChatGPT at all, the [ChatGPT Apps & Automations alternative](/alternatives/chatgpt-apps-and-automations) draws the line between a chat preset and a content engine.
Apps (via the Apps SDK, launched at DevDay on October 6, 2025) let third-party tools like Canva, Figma, Spotify, or Zillow run as interactive interfaces inside a ChatGPT conversation, built on the Model Context Protocol. Automations are scheduled tasks: a prompt you set to run once or on a recurring cadence, expanded in 2026 with web monitoring and integrations like Gmail, Slack, and GitHub. Together they are the two halves of building a repeatable workflow inside ChatGPT.
On July 9, 2026, OpenAI consolidated the App Directory into a Plugin Directory. A plugin is a package that can bundle one or more apps together with skills and app templates for a specific workflow. Existing connections kept working with nothing to resubmit, and the directory is available across ChatGPT web and desktop, ChatGPT Work, and Codex. It is the newer MCP-based system, not the deprecated 2023 plugins.
Not to your feeds. A scheduled task re-runs a prompt on a cadence and can monitor a page or draft text, but it is a reminder-and-monitor tool — roughly capped to about once an hour and unable to render branded video or carousels or publish to your social accounts. To make the media and publish it across platforms on a schedule, you need a content engine like Kompozy, whose Autopilot is a scheduled publish rather than a scheduled prompt.
The drafting and planning side, yes — invoke a design or research app inline, and schedule a task for a daily brief. But the last mile is missing: no finished on-brand media, no per-platform sizing, no persistent brand voice, and no publishing to your channels. Most creators pair ChatGPT for ideation with a content engine like Kompozy that generates the video, images, carousels, blog, and newsletter and ships them everywhere.