AI-Driven Hiring Process: Part 4 / The Interview Happens
Somewhere 'round step four, the system goes and does somethin' almost suspiciously restrained: it steps on back and lets two humans sit themselves down in a room — physical or virtual, camera on if we're lucky — and just... talk.
Ain't no AI joinin' this here meetin'. Nothin' gets pre-filled, summarized, scored, or nudged along. This is, on purpose, the one part of this whole eleven-step pipeline where the software's contribution comes out to exactly zero. Which, if you've been readin' this series, might just be the most radical design decision in the whole outfit.
The One Step That Refuses to Be Automated
It wouldn't be hard to go automate this. Plenty of tools'll happily transcribe the interview, sentiment-analyze the candidate's tone, flag "hesitation markers," and spit out a tidy little scorecard afterward, like some kinda fitness tracker for human judgment. We didn't build a lick of that in. Not 'cause it's technically hard — it ain't — but 'cause the interview's the one place in this whole process where the point ain't efficiency. It's contact.
Everything 'fore this step exists just to get a well-organized, well-documented candidate to this exact moment without nobody havin' to chase a PDF through six inboxes to get here. Everything after this step exists to turn a "yes" into a functionin' employee without nobody forgettin' a step. But this step, right here, is the actual hirin' decision happenin' in its native format: two folks, a conversation, and whatever gets communicated in them pauses that no transcript ever quite manages to catch.
What "Entirely Human" Actually Buys You
There's a reason this here matters more than it sounds like it oughta. A CV tells you what somebody's done. A structured record tells you what stage they're at. Neither one tells you whether this particular person, in this particular room, is somebody your team actually wants to work alongside for the next several years. That ain't no data problem. It's a judgment problem, and judgment — real judgment, the kind with intuition and discomfort and the occasional gut feelin' that's hard to justify on paper — don't currently, and arguably shouldn't, live inside a workflow engine.
So the interview stays deliberately unoptimized. No AI-generated question banks pipin' in live. No real-time "candidate sentiment: 72% positive" ticker runnin' in the corner of nobody's screen. Just the hirin' team, the candidate, and a conversation that either goes somewhere or it don't.
The System's Whole Job Here Is to Get Out of the Way
If Step 3's about capturin' everything so nothin' gets lost, Step 4's damn near the opposite instinct: capture nothin', structurally, and trust the humans in the room. The record just sits there quiet in the background with its NEW status, unbothered. No fields waitin' to be auto-filled. No AI hoverin' nearby, ready to helpfully summarize what somebody just said 'fore they've even finished sayin' it.
The interview happens. Afterward, somebody comes on back to the record and adds what they actually decided — not 'cause some form demanded it, but 'cause that's the next real step in the process. Which, fittingly enough, is the one where somebody's got to say yes formal-like, on paper, with their name attached to it.
Comin' Up
Next: what happens when a decision that felt plain obvious in the room's got to go survive contact with an actual approval process — and why "yes," "no," and "abstain" turn out to need three mighty different kinds of paperwork.