Shipped in November 2026: the publishing scheduler got the three upgrades working teams asked for (per-channel times, destination memory, and guards against scheduling in the past), the @auto agent harness entered beta - agent-assisted edits that land as reviewable versions - and AI shorts kept its human-first shape: the model proposes clip candidates, you pick. As always with these roundups, every item below is shipped, not "coming soon," and each one comes with its honest boundary.
Publishing: the scheduler grows up
Publishing through Buffer shipped last month (covered in its own post); November was about the details that decide whether a scheduler survives real use:
- Per-channel dates and times. One time for all accounts, or a different slot per account - the LinkedIn-morning, TikTok-evening pattern, in one scheduling pass. Choose "same time for all" or "different for each" and the form follows.
- Destination memory. The channels you published to last time are pre-selected on the next post - repeat publishing to the same destinations takes seconds, and adding a channel is one toggle.
- Past-time guards. The picker simply refuses to schedule in the past - past hours and minutes are disabled, and clicking an earlier day lands on a sane default time instead of an error. Invalid states became impossible rather than punished, which is how we try to ship all UX.
- Live delivery status. While a post is in flight, its state updates live in the workspace - scheduled → sent → delivered - without refresh roulette.
The loop these close
Approved version → scheduled to connected channels → live delivery status, all inside the project where the video lives. The approval record and the delivery record end up next to each other - which is the whole point of review-to-publish in one place. Publishing details: the publishing feature page.
@auto: agent-assisted edits, in beta
The @auto harness is now live in beta: point an agent-assisted pass at a task, and the result lands in your project as a new version - which means it goes through the same review gauntlet as human work. The agent does not publish, does not replace versions, and does not bypass review; it produces a candidate that a human approves or rejects on the review page, frame by frame.
Honest boundaries: this is beta. It shines on bounded, well-specified passes - assembling an approved plan, cutting candidates to spec - and it is not autonomous creative judgment. Agent editing is beta, and we would rather under-promise a feature that lands as a reviewable version than over-promise an autopilot. The broader agent workflow - including what stays human - is in the AI agents review workflow guide, and the MCP plumbing behind it in the MCP explainer.
AI shorts: candidates, not autopilot
AI shorts continues to work the way we think this category should work: the model surfaces clip candidates from a long video - the moments likely to work as standalones - and a human picks, orders, and reviews before anything ships. November's refinements are quality-of-life in that loop, not a new autopilot. The full method, including the failure modes to watch for, is in AI shorts from long videos.
What did not ship (yet)
So nobody reads this roundup as a roadmap: hosted Frooty Cloud MCP is still in development (the local Premiere and Resolve MCP bridges are the live path today), the After Effects panel work is ongoing, and automatic payment-triggered delivery remains a concept we have not built. We would rather list what is missing than let a roundup imply otherwise.
FAQ
Do I need to do anything to get these updates?
No - everything in this roundup is live on the product for existing workspaces. Publishing connects through your Buffer account from publishing settings; @auto and AI shorts appear in the same workspace you already use.
Is @auto included in every plan?
@auto is in beta and runs on AI credits rather than a plan gate; the beta boundary means its behavior is still settling and feedback shapes it. AI feature usage draws on credits - see the pricing explainer for how plans and credits relate.
Can the agent publish or send videos to clients?
No - agent-assisted passes land as versions in your project, and versions become links only when you send them. Publishing requires an approved video and your explicit scheduling action. The agent cannot cross the approval boundary; that is by design.
What's next for publishing?
The honest list: deeper post analytics stay in Buffer for now, and channel management remains on the Buffer side. We ship publishing features when they close a real loop, not on a content calendar - which is also why this roundup contains four shipped things, not ten promised ones.