WorkBuddy helped me maintain human review as the baseline for AI hiring tools

Saw the news this afternoon that Workday is being sued for age discrimination due to its AI recruitment tools. My first reaction was: do we, who use AI to screen resumes daily, actually have a safety net?

One detail in the news stuck with me: Workday's direct counter-attack strategy was to start from legal provisions, arguing that ADEA does not protect job seekers, completely avoiding the topic of algorithmic bias. This procedural defense strategy looks smart, but the case has already been approved as a nationwide class action lawsuit, which clearly indicates the court's attitude: hiring-related decisions made by AI algorithms will be regulated as employment actions. As an HR who uses AI for initial screening every day, I must admit it gives me chills.

All my resume initial screenings now run on WorkBuddy. I set up a workflow about ten days ago, and now every step has triggers for manual review. Today I'm not talking about efficiency, but compliance. Because I've indeed stepped into pitfalls and tested out solutions.

Let me talk about my initial configuration first. When creating a new project, I selected the "Recruitment" template, and the system generated three default areas: Resume Pool, Evaluation Sheet, and Interview Records. Then I imported all collected resumes into the Resume Pool, checked "AI Auto-Screening" in the settings, letting WorkBuddy do a round of keyword and skill matching based on job requirements, tagging them with A/B/C grades. A is strong match, B is potential, C is unsuitable.

After running about two batches of resumes, I found problems. A 48-year-old CFO candidate with over a decade of manufacturing finance experience was tagged as C by WorkBuddy simply because his resume didn't include words like "SaaS" or "Growth." The system wasn't wrong; it executed the screening criteria I gave. The mistake was that I hadn't set up a review process at all—a living person was rejected by a few keywords. I believe interviewers in the final stage likely wouldn't do this, but AI would without hesitation.

At that time, the Workday case hadn't made the news yet; I only realized something was off and went back to fix it myself. Now my WorkBuddy configuration is like this:

Open Project → Click "Automation" in the top right corner → Select "Screening Workflow" → Set my screening criteria as the first layer. The key step is here: Add a new node in the same workflow, select type as "Manual Confirmation," and set the trigger condition to "When candidate is Grade C." So now the flow is: WorkBuddy screens once based on the hard criteria I provided, sends all candidates tagged C into a "Pending Manual Review" list, and actively notifies me to check. Only candidates above Grade B proceed directly to interview scheduling.

This "Grade C requires manual confirmation" node is the insurance policy I think all teams using AI for resume screening should add.

Later, in the review area, I opened that candidate's resume, took a glance, and found a line in their work experience saying "Led the launch of a financial system for a manufacturing enterprise." Although this info didn't mention "SaaS," it is valid experience for the role. I manually changed her from C to B. All records are kept in the system, so if anyone asks one day, I can explain why the change was made.

There are two other details worth setting up. One is permission management: I set "Project Admins" to adjust screening criteria, while the "Interviewer" role can only view and add notes, not modify scores. Otherwise, if everyone can change screening criteria, the evaluation standards for the whole project will become chaotic. The other is regularly auditing review records. I export project logs every two days to see if there are many instances of Grade C being manually upgraded. If this happens continuously, it means the AI's screening criteria themselves are biased. Go directly to the "Job Requirements" page to adjust keyword weights and exclusion items, rather than fixing things one by one in emergency mode.

To be fair, if a company doesn't even set up a single manual review position, AI recruitment tools will only amplify the blind spots in their original decision-making. For those using WorkBuddy, at least build this review node first. Don't let algorithms unilaterally decide whether someone gets an interview. I'm not sure how the law will ultimately rule, but on my end, I've at least kept full decision trails and a mechanism for manual judgment supplementation. That day when I saw the Workday news, I opened the project log and flipped through it, feeling reassured.