I built a WeChat Official Account writing Agent that cuts work time by 80% — here’s the workflow filter and tool combo I used

Written with AI assistance. Figures without a traceable source were cut before publishing.

The same piece is posted as a thread on GitHub — that copy has a reply box under it, and this one does not.

My WeChat Official Account writing Agent now finishes its run in about 11 minutes of machine time. The same job used to take me 115 to 215 minutes by hand. I still review every piece at the end, which adds 10 to 20 minutes. Even with that review counted, my working time drops by 80 to 90 percent. The catch is that not every step deserved a machine.

A step goes to the machine only if it repeats, has a clear input and output, and needs no final human call. Research and drafting pass that test. Final edits stay human. The automated part runs on Codex paired with Workbuddy.

From Manual Drudgery to Scalable AI-Driven Workflows

I built a WeChat Official Account writing Agent that drastically slashes work time by 80-90%. Originally, the manual process demanded 115–215 minutes, but now the machine runs for just ~11 minutes, with an additional 10–20 minutes for human final review. This efficiency gain means you can redirect time from tedious content creation to more strategic tasks.

I’d argue that focusing on specific workflows matters more than broad productivity claims. Tools like KroWork allow you to fix common prompts into reusable AI applications, which effectively removes the need to manually regenerate outputs for every task. By turning repetitive actions into fixed intelligent agents, individual creators stop wasting time on redundant inputs and start scaling their actual output.

I was skeptical about AI agents at first, but I now believe we should evaluate them using the “three-node test”: repetitive tasks with clear inputs, outputs, and no need for final human judgment. Beyond simple chat, tools like Codex now automate complex workflows, from extracting PDF data to generating reports.

The 3-Question Filter for Agentizing Writing Workflows

To decide if a workflow node is fit for automation, apply the “3-Question Filter”. First, ask: Does this task repeat often? If tasks like content research or drafting pop up regularly, automation is a candidate. For example, in a scenario where a company needs to generate weekly blog posts about industry trends, the research for each post—drawing from similar data sources and focusing on comparable keywords—repeats consistently. This regularity makes automating the initial research phase a practical move. Second, is the input and output clear? Suppose you have a well-defined input, such as a set of specific product specifications for an e-commerce description, and a clear output expectation, like a formatted HTML product card. In such cases, automation can efficiently transform the input into the desired output. Third, does this task need final human judgment? Tasks that involve subjective elements, like tailoring a sales pitch to resonate with a specific target audience’s emotional triggers, require human intervention. Editorial decisions that shape the voice and tone of a piece also fall into this category. By using these three questions, you avoid over-automating and wasting costs. For instance, repeated content research and drafting are perfect for automation, but final edits should remain human. Let’s consider another angle: when applying the 3-question filter to a social media management workflow. The task of scheduling regular posts at specific times is repetitive, so automation can handle that. However, crafting the actual caption for each post, which might need to adapt to current trends or engage with followers in a personalized way, requires human creativity and judgment. This example highlights how the filter helps distinguish between tasks suitable for automation and those that need human oversight. Also, it’s important to know that not all tasks can be automated. For instance, tasks with ambiguous inputs or outputs, or those that demand immediate human decision-making, should not be automated. The 3-question filter acts as a reliable guide. You allocate automation resources where they make the most sense while preserving human involvement in critical areas

The practical application of this filter can be observed in several real-world cases where businesses have successfully integrated automation into their operations. For instance, an e-commerce company automated the process of updating product listings with seasonal discounts and promotions. By programming their system to adjust prices based on pre-set rules and timely product data, they managed to reduce human workload while increasing accuracy and responsiveness to market changes. This automation saves man-hours previously spent manually adjusting prices, thus allowing staff to focus on more strategic tasks such as market analysis and customer engagement This example shows that clearly defined rules and conditions for inputs and outputs are important for successful automation.

Reassess automation systems periodically to adapt to new market conditions or changes in business strategy. Automation systems aren’t set-and-forget. They need regular updates and tweaks to align with current business goals and market realities. Such reassessment could potentially involve altering the parameters within which automation operates or redefining which tasks are suitable for automation based on evolving business needs.

While the 3-Question Filter offers a framework for evaluating potential automation tasks, its success relies on continuous monitoring and adaptation. Automation contributes positively to operational efficiency and effectiveness, while including the critical human elements needed for specific decision-making processes. By maintaining this balance, businesses can use automation to boost productivity and innovation I buy the 3-Question Filter.

The Codex + Workbuddy Tool Combo

When choosing tools for AI writing workflows, Codex and Workbuddy have distinct strengths. Codex offers powerful AI capabilities, ideal for complex tasks like in-depth content generation. However, it has a steeper learning curve. Workbuddy, on the other hand, has an intuitive graphical interface. But Workbuddy’s AI capabilities are weaker, so it needs more manual adjustments. By combining Codex and Workbuddy, you use their strengths: use Codex for complex tasks and Workbuddy for simpler tasks and workflow management. This combo balances power and usability.

For instance, WorkBuddy’s user-friendly interface allows even those new to AI tools to quickly generate professional prompts by simply inputting plain language and clicking on prompts, which is particularly helpful for optimizing UI design. This means that someone with little experience in AI can still effectively use Workbuddy to refine their prompts Meanwhile, Codex’s skills and plugins come with a visual interface, which is one of the most beginner-friendly aspects of Codex, as it clearly shows users what skills and tools are available and how to use them, unlike some other tools that can be confusing to navigate. This visual clarity in Codex helps users understand and use its advanced features more easily.

When combining these two tools, the benefits extend beyond just usability For example, a content creator working on a detailed blog post can use Codex to generate in-depth sections that handle complex ideas with its powerful AI Simultaneously, they can use Workbuddy to manage the overall workflow, such as organizing research materials and ensuring the structure of the post is coherent. This integration addresses both the depth of content and the efficiency of workflow management. The combination of Codex’s AI capabilities and Workbuddy’s intuitive interface creates a setup that allows users to tackle a wide range of tasks with greater ease and effectiveness.

Workflow Implementation and Efficiency Results

The Agent handles content research, drafting, and even image generation, while humans focus on final review. The machine runs for about 11 minutes, and human review takes 10–20 minutes, leading to a total time reduction of ~80%–90%. For example, tasks that once took hours can now be done in a fraction of the time. Tools like Doubao offer AI image and video generation features that support high customization for WeChat OA content. You can describe specific scenes, like “a morning street with warm tones and a sense of hustle,” and the AI will generate corresponding visuals. This integration speeds up the entire content creation process.

I’d argue that the true power of AI in product selection isn’t just about raw speed, but about how it handles sheer volume. According to the insight, AI can analyze thousands of material items in a single day, performing at a rate over 5 times more efficient than human labor. This shift changes the economics of market research. When you are monitoring e-commerce trends, you no longer have to manually sift through endless feeds to find potential winners. Instead, the machine processes the data, leaving you to handle only the final review. Relying on this approach allows businesses to remain agile without burning out their teams on repetitive data sorting. While some argue that automation might miss the nuance of a human eye, I find that for high-volume tasks like trend spotting, the trade-off is clear: the AI does the heavy lifting of processing thousands of items, while I retain the authority to make the final call on which products actually fit the brand’s direction.

Key Lessons Learned and Best Practices

One key lesson is to avoid over-automating by focusing only on nodes that truly need it. Multi-Agent systems are for parallel execution, not isolating tasks, so using them unnecessarily increases costs. Another best practice is to start with free tiers of tools like Doubao to test capabilities before upgrading. The free tier is often enough for daily use, and premium plans (around 68 RMB per month) are available for advanced features. Finally, be ready for manual adjustments when using Workbuddy, as its AI needs more human input during setup. Following these practices optimizes your AI writing workflow.

Also readable on Telegraph.

Read next

All 68 write-ups

The 12 figures in this piece — each with the sentence it came from — are in the figures table, alongside 632 more, as JSON and CSV.

Topics: Automation Systems · AI Automation

Part of llm-api-pricing — field notes on AI coding agents.

Did this save you an afternoon? A star on the repository is the whole ask — it is what puts these in front of the next person looking; the data is CC BY and does not require starring.

One thing this piece could not settle: whether final review shrinks with practice. The last time you let an AI draft a WeChat Official Account post, how many minutes did your own final edit take? Reply with one number, from memory. The reply box is on the thread copy of this piece.

Want a figure that is not in here yet? Say which metric, which provider, which unit — in one line. One required field, and the page you came from is already filled in. Got a better number? Open an issue — that form knows which write-up you came from too; corrections and counter-data are the point.