Using WorkBuddy, I Found That as AI Resume Tools Get Better at Packaging, HR Needs Evidence-Based Processes

I disagree with the current narrative in job hunting guides that the most important value of AI resume tools is helping supplement keywords, quantify achievements, and pass ATS. From an HR perspective, this statement is only half true. What truly gives me chills is that when more and more resumes can rewrite "participated" as "led" and "looked at data" as "improved retention," screening doesn't become easier—it becomes more expensive. Because I have to spend more time judging authenticity, and guard against rushing to let the wrong person into the interview room.

Recently I took on a Content Operations role. The position itself wasn't complex, but the urgency was high. The business side wanted lists, interviews, and feedback. Previously, this kind of task was the most feared because resumes came in from WeChat, Feishu, email, and cloud drives; interviewer feedback was scattered across Teams and group chats; and job status lived only in my head. That day, I decided to move the entire pipeline into WorkBuddy. I created a project called "Content Ops - 2026 Q3." I set permissions in three tiers first: hiring leads could modify JDs and screening criteria; interviewers could only view candidate evidence summaries and interview records; external channels couldn't see internal evaluations. This step seems boring, but it's truly useful. Connecting tools is just the first step; if permissions and classification aren't done well, the more automated it gets, the messier it becomes.

After receiving resumes, I didn't let AI screen candidates directly. In WorkBuddy, I first used Skills to write pitfalls encountered over the past three weeks into a checklist: Is there independent ownership? Are there data metrics? Is the cycle and resources explained? Is there post-failure handling? This Skill wasn't ready-made; I built and refined it as I went. Previously, I tried to state requirements clearly all at once, resulting in smooth recommendations that felt increasingly shaky upon closer inspection. Later I learned: let it do one thing at a time. JD breakdown only breaks down the JD; experience evidence only looks at evidence; risk review remains a manual node.

Here is a counter-intuitive judgment. Many current AI resume tools optimize "participated in project" to "Led feature iteration, resulting in 18% user retention increase."

For job seekers, this sounds great. For HR, this is dangerous. Because the more resume language resembles product reporting, the more I need to probe for underlying evidence. WorkBuddy's help here isn't generating prettier sentences, but breaking apart the pretty talk. I ask it to extract "what the candidate claims to have done" from the resume, then separately generate "what needs verification." For example, if a resume says "Led campus event covering 5,000 people," WorkBuddy breaks it into questions: What was your role? Where did the budget come from? Is the data registrations or actual attendance? Do you have retained retrospective reports? These questions aren't necessarily asked every time, but interviewers at least don't have to ask based on gut feeling.

I also took detours. I had just started using Single Agent mode less than a week ago, wanting to save effort by having it complete initial screening, ranking, and recommendation all at once. As a result, it pushed a candidate with a beautiful template but only "assistive" experience to the top, even writing a comment like "possesses strong user growth capabilities." My first reaction was to stop. It wasn't that the model was stupid; I had set the boundaries wrong. Recruitment isn't copywriting; one ranking error wastes half an hour of interviewer time. Later, I downgraded Single Agent to information extraction and handed formal recommendations to the Expert Team. My tests show the Expert Team is slower than Single Agent—for a batch of 42 resumes, Single Agent takes about 8 minutes, while the Expert Team takes about 15 minutes—but manual rework decreased significantly. Of the 6 people who made it to interviews, 3 were missed by the initial Single Agent version because their resumes weren't as "pretty," but their experiences contained real actions.

Speaking of "pretty," I recently looked at Teal and ZhiWuKong. I used Teal for two weeks; it's indeed suitable for helping candidates tidy up their experiences. I also used ZhiWuKong for two weeks; job recommendations and resume matching are intuitive. They solve "how do I get seen by the system." WorkBuddy solves something else: how recruiters avoid being deceived by the system. This difference is huge. The better AI resume tools get at filling gaps, the more HR needs to turn processes into chains of evidence.

WorkBuddy solves not whether the resume looks good, but whether the recruitment process has a chain of evidence.

WorkBuddy is not a chat box, it's a recruitment pipeline

When the chain of evidence lands in daily operations, it's actually quite fragmented. I've just started using Group Bots for a few days and am still testing. Initially, it kept showing "connecting." I checked the API Key, found permission scopes weren't fully enabled, and the callback URL had a typo. Used for less than a week, not yet stable, but basic issues can be handled automatically. Candidates ask "Is the job still open?", "Can you send me the JD?", "When will interview results be out?" The bot replies with templates first, transferring complex questions to humans. Previously, this annoyed me the most—"Dear, this position is closed," yet they still ask. It's not rudeness; it's lack of information maintenance. If job status relies on HR memory, it will eventually hurt people.

Beyond Group Bots, I gradually connected shared tables, cloud drives, WeChat, and Teams into WorkBuddy. I've used WeChat, Teams, shared tables, and cloud drives for less than a week each, mainly for messages, meeting records, candidate stages, and portfolios. Here I stepped on a pitfall: initially, cloud drive permissions were too broad. Business sides wanted to view candidate portfolios directly, resulting in a candidate uploading a file containing personal contact info. Later, I changed it so WorkBuddy projects only contain desensitized summaries; originals require application to view, leaving a system trail. This isn't a big deal, but it's typical. Automation isn't opening the door; it's placing the nameplate, keys, and surveillance in the right positions.

I've tested Workday for less than a week here, because seeing news about AI recruitment tools being sued kept me anxious. WorkBuddy doesn't intend to replace Workday, but I'm now more wary of any automatic screening. I set a bottom line for myself: any elimination reason generated by WorkBuddy cannot be sent directly. It must undergo manual review, especially involving words related to age, education, marital/childbearing status, region, or school background. Once, it listed a candidate's career gap as a risk point; I immediately deleted that rule. Gaps can be asked about, but cannot be default reasons for filtering people out. This feature is useful for those standardizing processes, but a landmine for those seeking shortcuts.

From my experience, WorkBuddy is best suited for: teams running multiple parallel roles, non-fixed interviewers, data scattered across systems, and those needing audit trails. It's not suitable for those who just want one-click JD generation, nor for those without real standards. If you haven't thought clearly about what the role actually requires, WorkBuddy will only amplify your ambiguity. It's like an honest mirror: if your process is messy, it gives you mess; if your standards are clear, it gives you speed.

Did this Content Ops role succeed? Yes. Not because it was perfect, but because WorkBuddy placed job status, resume evidence, interview feedback, and Offer follow-ups on the same timeline. The business side no longer asks "Where is the person?", interviewers no longer dig through chat logs, and candidates aren't repeatedly pestered by unmaintained groups. Recruiting is hard; often it's not that people aren't excellent, but that no one in the organization takes responsibility for the process.

If you're recently tormented by recruitment or collaboration processes, don't first ask if AI can help you write; first use WorkBuddy to build a minimal closed loop.

One real project, three nodes, one review rule. Nodes can be job standards, resume evidence, interview feedback. The review rule can be simple: any automatic screening result must be manually confirmed before proceeding to the next stage. Run it for a week, and you'll find that what you save isn't just time, but the peace of mind of finally knowing where things are stuck.