An Agent That Watches Our Blog and Drafts the Social Posts

Publishing a blog post is the easy half. The other half is telling people about it: a LinkedIn post, something shorter for X, an Instagram caption, and if you run country pages, the Italian variants on top. That work is small, repetitive and endlessly postponable, which is why it reliably did not happen.

So in August we taught our own product a new trick. Engage Social, our social publishing tool for SAP Sales and Service Cloud V2 customers, now has a blog watcher: a small agent that notices when a new article appears on a blog and drafts the announcement posts, one per platform and language variant, straight into the content calendar as drafts.

TL;DR: Our blog watcher in Engage Social notices every new article and drafts the announcement posts per platform and language, as drafts only. A person reviews, edits and schedules them. The agent removes the blank page; the judgement about what goes out stays with people. That split, agents draft inside a human workflow, is how we approach Business AI generally.

The concept: agents draft, humans publish

The design decision that matters is not technical. The agent has no publish button. Everything it produces lands as a draft, visibly marked as such, waiting in the same calendar where a human plans everything else. Someone reads it, sharpens a sentence, moves it to a better slot, and only then does it go out.

We think this split is the honest one for marketing content today. The blank page was the real bottleneck: nobody enjoys writing the fourth variant of the same announcement. An agent is excellent at fourth variants. What it cannot judge is whether this particular post lands right for this particular audience this particular week. That judgement stays with a person, and the tool’s job is to make the judging take two minutes instead of forty.

Platforms are not copies of each other

The second lesson came fast once real drafts appeared. A good LinkedIn post is a bad X post; an Instagram caption is its own animal; and the Italian version for the Italian company page is not a translation, it is a retelling. So the agent drafts per platform and per variant, and we stopped pretending one text fits all. Less is more still holds, just not for the number of variants: it applies to how much of each variant a human still has to touch.

And because we run Spadoom’s own channels through it, we use it ourselves every day. Every rough edge in the drafts is our problem before it is any customer’s. That feedback loop improved the agent faster than any specification would have.

The quiet takeaway

The fashionable version of this story would promise a marketing department run by AI. Our version is smaller and, we think, more useful: one well-placed agent removed the most postponed task in our content routine, and the humans got their time back for the part that actually needs them. That pattern, agent drafts inside a human workflow, is what we currently believe in for Business AI, in marketing and in CRM alike. It is the same principle we apply to SAP’s own agents in Sales Cloud V2, which we describe in agentic AI in SAP and what it means for sales teams.

The best test: this very article arrived in our social calendar as four drafts, written by the watcher, a few minutes after it went live.

The same discipline applies to any agent that drafts on your behalf: a named owner, a review step and a measurable baseline. That is how we scope SAP Business AI work too, and SAP’s quarterly Business AI release highlights set out what is available in the product today. Running our own tools on our own business is a habit, not a one-off: we also run SAP Sales and Service Cloud V2 for Spadoom itself. Discuss a use case.