19 articles, 1,800+ new followers: how she turned PilotDeck into an automated operations expert
Xiaoshan, a developer in the open-source community, recently ran an experiment in growing a WeChat Official Account from scratch. With no extra promotion, she published 19 articles in a row, gained more than 1,800 new followers, and went on to receive reader inquiries and partnership invitations.
The numbers come from her own records. They don't guarantee that anyone who follows the same steps will get the same result, but they point to a concrete question: individual developers can use agents to turn easily interrupted content work into a process that runs again and again.
The tool Xiaoshan used is PilotDeck. She does not treat it as a writing tool alone. Her focus is on controlling the entire operations chain: let the agent take over the time-consuming, repetitive work that still has to be done, and keep her own time for judging topics, choosing sources, and sharpening arguments.
First, get AI out of the chat box
Many people have tried asking AI to write for a WeChat Official Account, and what they get is usually a fluent article that still reads like a template.
The reason is that running an account is ongoing work with many steps. An AI chat box can write a passage of text, but it struggles to remember long-term rules, and it won't automatically pick up the image selection, layout, and publishing that come next. What we actually need is a working environment that can hold long-term context, execute across steps, and keep accumulating experience.
Within PilotDeck's architecture, Xiaoshan set out to turn account operations from one-off content generation into an agent workflow that runs continuously. Three efficient modes of human-agent collaboration took shape:
Mode 1: minimal oversight — a single sentence runs the entire pipeline
For a long-running project with many subtasks like account operations, you can create a dedicated WorkSpace in PilotDeck and open different conversations inside it to write different articles. Within one WorkSpace, effective interactions and execution traces are retained as reusable experience that keeps evolving.
When you start a new conversation in the workspace, past memories and Skills are called up automatically, and the entire topic selection → research → drafting → images → layout → save to drafts pipeline can be fully automated.
So in a new conversation, Xiaoshan only has to state one request.
"Find me a hot new topic, write an article for the account, and save it to drafts."
The agent then draws on the memories and skills built up in earlier conversations to handle topic selection, drafting, and cover image generation. All Xiaoshan has to do is review and publish from the drafts folder in the account's admin console.
This article includes video content. Please view it in the original WeChat post.
Mode 2: heuristic co-creation — deep back-and-forth between human and agent
When a topic interests her, Xiaoshan first talks it over with the agent in PilotDeck for several rounds.
She presses on where the material comes from, asks for angles that are easy to miss, and points out directly where an analysis is not solid enough.
Once the discussion is done, she has the agent turn the material into a long-form article suited to how people read on the account, and save it to drafts.
The article still needs human review, but the focus shifts from fixing sentences to judging whether the facts, logic, and argument hold up.
Mode 3: long-horizon research — leave it running and get a deep report
Conventional AI chat tools tend to jump to conclusions, and their answers often stay on the surface. In professional settings, what we need is an agent that can reliably carry out long-horizon tasks for hours and produce well-grounded, dependable results.
When Xiaoshan hands this kind of task to PilotDeck, the agent first proposes a research hypothesis, then calls tools to search, cross-checks information, and keeps digging when something looks doubtful. The task can run for hours on end and end in a whitepaper or research report tens of thousands of words long.
In the meantime, Xiaoshan can go do something else. When the report is done, she comes back to check the citations, the reasoning, and how trustworthy the conclusions are. Once the material checks out, all that's left is to split the report into several articles and run it as a series on the account.
How the end-to-end automation works
An agent that only writes drafts can hardly lighten the load of running an account. In this workflow, Xiaoshan also connected images, layout, and saving to drafts, closing a genuinely end-to-end loop.
Beyond topic selection and content generation, PilotDeck can call image generation APIs to produce header images, thumbnails, and in-article visuals according to an established visual spec. After the article is finished, it formats the piece to the account's layout requirements and can save it straight into the WeChat drafts folder in the backend.
Xiaoshan no longer copies things back and forth between tools. She spends most of her time reviewing and revising, with one final human check before publishing.
The process gets smoother the more it is used, because memory and skills are continuously updated and retained.
Whenever she finds a new layout rule, a better writing prompt, or a more effective search method during operations, Xiaoshan adds it to the relevant Skill. Previously published articles stay in the project memory in PilotDeck, and user preferences confirmed in conversation are recorded automatically.
The next time she opens a new conversation, the agent already knows what this project has written, which conventions the articles follow, and where extra care is needed. Xiaoshan does not have to restate the background and requirements from scratch, and the agent can pick up where it left off.
From producing knowledge to monetizing it
Over time, a clear feeling took shape as Xiaoshan used the system: once an article is published, the feedback it generates leads to the next round of topics.
A question starts with searching for material. After reading it, she has to reorganize the information and confirm that she has really understood it. Writing the article forces the vague parts into the open. After publication, reader comments, inquiries, and partnership conversations bring new questions.
The 1,800-plus new followers from 19 articles gave Xiaoshan two concrete things. She learned which content people care about, and readers with real needs found her. The inquiries and partnership invitations are the direct result of that connection.
For individual developers, this process matters. Sustained research gradually sharpens professional judgment. Only by writing those judgments down can others know what you are working on. Reader feedback then sends the next question back, and the content and the person's ability grow together.
An agent can take over a large amount of repetitive, time-consuming work. Search, image assembly, layout, and saving to drafts can all be handed to it. What people need to do is decide which question to research, which material to trust, and what the article should ultimately say.
This division of labor also determines how far an article can go. An agent can help one person finish an article faster, but whether the article lasts depends on the author's understanding and judgment of the problem.
In the AI era, the human role is more than a "prompt pasting tool," and AI should not be left to churn out meaningless internet junk.
With PilotDeck, Xiaoshan turned an agent from a "chat box" into a digital partner that keeps evolving. If you want to build your own automated agent workflows, come explore with the OpenBMB community!