Job Platform Rankings Show the Entrance; I Use WorkBuddy to See the Exit
Community Discussion · General

Job Platform Rankings Show the Entrance; I Use WorkBuddy to See the Exit

A DeerA Deer1d ago2026/11/17 161 views

A couple days ago I came across an article about how to choose job-hunting platforms for operations people transitioning to AI business roles in 2026—Maimai, Boss Zhipin, listed one after another, each with a line about who it suits. My first reaction was a bit out of place: this ranking is about choosing an entry point for job seekers, but for someone like me sitting on the screening side, it's actually telling me what kind of resumes I'll be receiving next.

The same batch of material had an even more pointed line, from a Tencent Cloud article about the 2026 AI job market—the gist being that the rule has shifted from "can use tools" to "delivers results," that proof of ability has to rest on verifiable things, and that resumes should be written in the language of results. I agree with that. In the past two months I've interviewed about thirty operations people wanting to move into AI business roles, and what really holds them back is the resume itself—full of "responsible for," "participated in," "assisted with," not a single result you can reconcile against.

So I built an A/B comparison with WorkBuddy. Side A is the platform dimension—where you applied from, what role you applied for; side B is the results-language dimension—what you actually delivered. Comparing item by item, side B is the column that decides whether you get an interview.

First, have it break the JD into reconcilable fields

My setup is an ordinary office laptop, using both the WorkBuddy web and desktop clients. This won't necessarily apply to everyone, especially if your JD itself is written vaguely.

Open WorkBuddy, click New Task on the left, and choose task type Document Processing. The first thing I throw in is the role understanding, not the JD text. Last week I wrote something called a role understanding card—I hand-wrote that card first, then had the hiring manager poke holes in it and revise it once. This step can't be skipped; skip it and everything after is empty. Upload the card and the JD together, and write the instruction in the input box. My instruction template is fixed like this: extract five columns from this JD—verifiable deliverables for the candidate, basis for judgment, corresponding business stage, missing items, whether to recommend for interview. For output format, choose Table in the bottom right—don't choose plain text, or you'll have to format it yourself when it comes back.

The key is to hard-code the field names. The first time, to save effort, I just said "analyze what capabilities this role requires," and what came back was a long sprawling block that looked plausible but couldn't actually be aligned with resumes. Once the fields were fixed to five columns, the table could be used directly as a sieve. I saved this instruction into Common Instructions, so next time I create a task I can pull it up from the dropdown.

Once resumes come in, fill them in reverse using the same fields

The second step is to batch-drop the resumes in. Don't do more than twenty at a time—I tried thirty together once, and several scanned PDFs were recognized all garbled, with names and company names running together. Convert scanned files to text versions before uploading; I stepped in that pit once and don't want to again.

The instruction is still those five columns, just replacing "whether to recommend for interview" with "gap description." When it finishes you get a table: on the left, the deliverables the role requires; on the right, what this person actually wrote. The empty cells are where you need to probe in the interview. This step takes about ten minutes, much faster than reading them one by one myself, but it replaces hands, not judgment. After the table comes out I still randomly pull ten to review myself—and when sampling, I can't just pick the familiar ones; it has to be forced random, or whatever gets missed stays missed forever.

For collaboration I split it like this: the output table goes into a shared folder, the hiring manager gets read-only, no edit permission—if they have comments they add annotations on the side, and I consolidate and revise. I've been through the result of giving edit permission once: three people editing the same table at once, the field names got mangled, and later runs couldn't align.

For maintenance, just one rule: every Friday afternoon run through that week's resumes, then archive the table with the date in the filename. If you don't archive, two weeks later even you can't tell which is the latest.

Back to that platform ranking. Maimai suits people who want to get into big companies and AI firms—true enough, but it can't do anything about whether there's substance in your resume. The platform decides who sees you; results language decides whether you stay after being seen. If you're transitioning roles, I suggest not rushing to switch platforms—take the work you've done in the last three years and fill it in yourself across three columns: deliverables, basis for judgment, business stage. The rows you can't fill are what you need to work on over the next six months. Same on the HR side: those requirements in the JD that can't be written as deliverables probably won't land even if you hire for them.

2 replies

?
Ctrl + Enter to reply
Weiwei's Shop

Running thirty at once just serializes them. I wouldn't even cram a packing-and-shipping batch like that. I'll own this pit—switch to the text version first.

Pixel Dust
Pixel Dust18h ago

The detail about "force a random sample of ten" is so real. If it's not random, the ones you end up sampling are all the ones you're familiar with, and the misses stay missed.