Interviewing and Teams
Drake Kelly & Olive
Interviewing and teams: the loop, the panel, and whether the people in the room agree on what they saw.
The interviewing and teams beats: the loop itself, the questions that survive AI coaching, the panel that reads one answer four different ways, and what shows up in month one when the interview was the only evidence. Where the two meet is calibration, and the question it keeps returning to is what an interview is worth when the agreement behind it is thin. It offers no scorecard that ranks a candidate, and no verdict a panel could not defend out loud.
121 articles
On this beat
- What the Record Shows About AI Interviews, Accents, and Disability One major vendor scores the transcript, not your face or voice, but speech recognition carries a measured accuracy gap. What is scored, what is documented, what to ask.
- Calibration Is a Standing Session, Not an Onboarding Slide Calibration has one measurable output: how far apart raters land on the same submission. Track the spread across sessions, and hunt the discount nobody wrote down.
- The Recruiter Chairs the Debrief and Guards the Evidence The recruiter chairs and does not vote, the hiring manager decides, and the chair's job is making every claim in the room say where it came from.
- Three Failures, Three Deadlines, and Only One Is the Hire's Fault A skill gap, a judgment gap and a badly drawn role look identical at week six and run on different deadlines. How to tell them apart before the clock starts.
- How Long They Keep Your AI Interview Video, and How to Get It Deleted Illinois gives a 30-day deletion right for an AI-analyzed interview video. California gives a narrower one. Everywhere else, the video waits out a retention schedule.
- If You Barely Use AI, Say So and Bring the Judgment Usually not a dealbreaker. Describe where you have used AI, where you decided not to, and what you would verify before trusting it in your field.
- Week One Buys One Thing: An Unrehearsed Look at How Someone Works A first week buys one thing later weeks cannot: an unrehearsed view of how someone works. What to schedule, what to compress, and what week one cannot tell you.
- Build the Reps Your First Job No Longer Hands You AI now does the cheap-to-fail tasks that taught junior judgment. Replace them: attempt the task yourself, then note where an assistant's answer diverged from yours.
- The Manager's Screen Buys the Spec, Not a Second Opinion Run the manager's twenty minutes as an extraction of the hard part of the role, then watch what the candidate asks about it. A second opinion is the cheap output.
- Forecasts Missed Both Ways. Track Your Own Task List. No forecast on job timelines has a good record in either direction. Here is the quarterly task review that updates your plan on evidence instead of the next headline.
- An Intake Meeting Produces a Decision Rule, Not a Wish List The deliverable is one page the manager has corrected: outcomes, requirements mapped to stages, the evidence each stage produces, and who breaks a tie.
- Interview Copilots: What They Cost You in the Follow-Up Tools like Parakeet and Final Round feed you answers in a live interview. The real cost is a suggested answer collapsing under a follow-up you have no model for.
- Plan Three Questions and Nine Follow-Ups, Not Twelve Questions Budget three or four planned questions with two or three follow-ups each. Go down rather than sideways, stop when the answer turns general, and track who got the depth.
- Name What Each Round Measures, Then Delete Everything Else Write one sentence per round saying what it establishes, then cut every scorecard line that is not it. No programme, no disclosure, no budget required.
- One Cell Per Round: The Coverage Map That Ends Repeat Rounds Competency labels do not separate two rounds. Map competency against evidence type, give each round one cell, and the repeats become visible.
- Your Interviewer Is Reading AI Questions. Answer the Script. A script read verbatim is usually a structured interview, the format with the stronger validity evidence. The real tell is whether follow-ups engage your answer.
- Prompt Engineering Went Into Every Job, Not Away The prompt engineer title mostly stopped appearing in postings, but the work behind it moved into analyst, support and marketing roles. Train for the role, not the title.
- AI Is the Reason Given. It Is Rarely the Whole Reason. An AI-attributed layoff can mean real automation, a bet that has not landed, or framing for investors. The tell: whether your old role returns to postings.
- The Internal Move Needs a Shipped Thing, Not a Course That evidence is a small working thing someone else already uses, not a certificate. Ask the receiving team its bar, and raise backfill before it stalls the move.
- Output Ramps in Weeks Now, Judgment Still Takes Months The eight-month ramp figure is HR self-report from a 2012 online survey. Carry two numbers instead: time to usable output, and the week you stopped checking the work.
- A Veto Is Fine If It Has to Name What It Saw When the panel disagrees: the manager decides, anyone can block, and every position names what was said or done and in which round. Uncited vetoes do not count.
- Ask the Two Questions Only Someone Who Read the Role Would Generic closing questions are instantly recognizable now. Build one from an ambiguity in the posting and one from what was actually said in the room instead.
- What to Ask an Employer About the AI in Their Hiring Process Eight questions that get real answers instead of a defensive non-answer, split into what AI does in the job and what it does in the hiring process.
- Can You Turn Down the AI Interview and Stay in the Running? Often yes, but not automatically: ask for an alternative format first, and know that declining with no other path usually ends the application.
- Use AI for Company Research, Then Check Three Sources A model answers confidently from stale training data. Use it for questions about a company, then check anything you plan to say aloud against three primary sources.
- Calibrate on Excerpts and a Disagreement Log, Not on What a 3 Means What a calibration session should do: put one recorded answer in front of every interviewer, collect the evidence each one names, and log the competency that split them.
- Lock the Scores Before the Debrief, and Keep the Split Visible A rating filed after seeing a colleague's is a vote, not an observation. File first, check whether the tool hides submitted cards, and treat a split as the result.
- Seniority Prices Judgment, and Judgment Just Got Scarcer Watching a junior match your output in minutes is not proof your experience stopped mattering. The throughput half is what compressed; the judgment half got scarcer.
- Switch for the Task Mix, Not for AI in the Job Title Career-switch marketing sells a destination title, not your risk. Compare task exposure on both sides of the move before you pay for a bootcamp or a certificate.
- Show the Error You Caught, Not the Output You Shipped Keep a record of the times you caught AI being wrong: what it said, how you found out, what it would have cost. That is the evidence employers ask for.
- One Interviewer, Two Passes: Split Evidence From Verdict A single interviewer can reproduce what a panel actually did: write evidence with no ratings, score it the next morning against a bar set in advance.
- Why Your STAR Answers Sound Like Everyone Else's Now STAR answers stopped standing out because a model fills the template in seconds. Keep the structure, cut your story count, and rehearse for depth instead of phrasing.
- Lock the Scorecards Before the Transcript and the Debrief Submit before you read another card, the transcript, or its automatic summary. Sequence is what protects the account, and a 24-hour deadline cannot see it.
- Say What You Fixed, Not What the Model Wrote A three-beat answer for the AI-use interview question: the task you handed over, what came back, and the specific thing you changed and why.
- The Manager Writes the 30-60-90, Starting From the Loop's Open Question A 30-60-90 plan written by a candidate to win the job is a guess. What belongs in each column, who writes it and when, and the field every template leaves out.
- Interviewer Training Ends in a Rated Practice Round, Not a Deck Interviewer training that works ends in a rated practice round: two completed write-ups read, one recorded round scored cold, and a write-up a second reader can act on.
- Recognition Error Is Not Evenly Spread. Do Not Score the Transcript A vendor accuracy figure is an average over the speakers they tested. Measured error is far higher for deaf and disordered speech, so keep the audio.
- Preparing With ChatGPT Is Allowed. Using It Live Is Not. Rehearsing interview answers with ChatGPT beforehand is normal preparation. Running it live during the interview is a different act entirely, and here is the actual line.
- What an AI Interview Scores, and What It Does Not One large vendor publishes what its AI interview scores: the words in your transcript, not your face or voice. How to find out what yours measures.
- Four Kinds of AI Interview, and What Each One Records An AI interview can mean four different things. Here is what each format records, who reads it, and the one thing worth asking before you sit for it.
- A Good Question Is One a Prepared Stranger Cannot Answer A good interview question turns on something only that candidate did, and makes strong and weak answers visibly different. Four checks for the bank you have.
- A Scorecard Holds Evidence, a Call, and Nothing Else A hiring scorecard needs five fields: the claims the round tests, the evidence for each, where the observation came from, the claims it missed, and one call.
- Cut the Round That Has Never Changed a Decision Cost picks the wrong round. Cut the one whose verdict has never differed from the round before it, then label what remains a gate or a sell.
- Every Interviewer Owns One Competency or Comes Off the Loop Staff the loop from the competencies you cannot learn anywhere else, give each seat one and name the evidence it owes, and remove the seats that duplicate or ratify.
- Put the Hiring Manager in the First Pass, Not Only the Shortlist Recruiters should own the pass, the pace and the record. A manager's hour buys more spent writing criteria and reading rejected applications than grading a shortlist.
- AI Is Taking Tasks From Your Job Before It Takes the Job The research on AI and job loss operates at the task level, not the job level. How to size your own exposure instead of guessing from a headline.
- Should You Add an AI-Fluency Round to the Interview Loop? A standalone AI round tests prompt trivia. Put the observation inside the round you already run, on the AI-assisted task the hire will do in their first month.
- Advertising AI Tool Access and Rules in the Posting Name the tools, the plan tier, who pays for the seat, and one real restriction. What experienced AI users read a posting for, and what an AI-forward line tells them.
- Buy the Certificate Only When Someone Outside Asks for It Buy a credential only where an outside party demands one. Otherwise the same money buys protected hours on real work and a senior reader at the end of them.
- Should AI Be Its Own Competency in the Framework? Fold AI into an existing competency where it changed the method, and give it its own where it changed the deliverable. The split runs function by function.
- Why Do Candidates Who Demo Brilliantly With AI Fall Apart in Month One? A demo rewards a finished artifact made under supervision. Month one bills for the checks nobody watched, in four patterns, each with the question that catches it.
- What a Good AI Answer Sounds Like at Junior, Mid and Senior Junior answers show verification. Mid-level answers show delegation. Senior answers show a review rule and what it cost. Write the three bars before the loop opens.
- Can the AI Fluency 4Ds Work as an Interview Rubric? The four Ds name what to watch, not what counts as good. Anchor each one to your occupation's own tasks, score them separately, and never total them.
- AI-Written Interview Questions Are the Ones Candidates Already Practiced Drafting questions with a model is fine. Keeping the ones it can answer is not. The stress test that sorts a generated list, and the two questions to write by hand.
- Are Candidates Using AI During Live Interviews? Yes. Real-time interview assistants are shipping products, strongest on questions with one right answer. Ask about decisions the candidate personally made.
- Ask Every Participant Before the AI Notetaker Joins the Call Ask every participant before the bot joins. The count of all-party consent states moves, the strictest participant can govern a panel call, and a banner is not consent.
- Can an AI Notetaker's Summary Go in the Hiring File? Not that sentence. Demeanor is an inference, not a record. Strike the line before the file closes, keep the transcript, and write what was asked and what was answered.
- Prep With AI Is Not the Thing Your Rule Is About Preparing with a model is coaching, and coaching was never bannable. Write the rule about live assistance, then fix the question that made prep feel like a problem.
- The AI Skills Gap Is a Specification Problem First Write the skill as observable behavior for two or three roles, then look at real work. Most teams find a distribution they had not noticed rather than a shortage.
- Why Does an AI-Skills Hire Work Like the Rest of the Team After Six Months? The hire wasn't wrong. Six months taught them your real standard, and no review step ever asked for the framing, the source or the check they were hired to do.
- How Do You Judge Communication After an AI-Translated Interview? Judge what came through: the decisions, the evidence demanded, the answer that changed under a follow-up. Require unmediated English only if the job does, in writing.
- How AI Use Belongs in a Performance Review, and How It Doesn't AI use does not belong in a review as its own rating. Grade the work, and write the part of the job AI changed as observable behavior on competencies you already have.
- Illinois Consent Is the Easy Part of AI Video Interviews. BIPA Is Not. The Illinois consent form is cheap. The biometric statute behind it carries a private right of action, so find out first whether the tool builds a template.
- Apprenticeship, Internship, or Rotation: Which Survives AI? Match the shape to how long your field takes to make someone useful: rotations for long ramps, apprenticeships for regulated ones, internships as short paid trials.
- Do Assessment Scores Still Predict Performance With AI in the Workflow? Validity is measured against one job's performance data, so re-validate per role: the tests that measured production speed lost the most when AI arrived.
- Behavioral Questions Still Work If You Ask for Events, Not Stories The rehearsed arc was never the signal. Ask for a date, an artifact, the person who disagreed and what it cost, then score how much the answer narrows.
- How Do You Get an Interview Panel to Judge AI Use the Same Way? Three acts, anchored in the hiring role's own vocabulary, scored alone and in writing before the debrief opens. The one-hour calibration that ends five private standards.
- How to Build an AI Upskilling Program That Changes the Work Three layers: tool access, supervised practice on the team's own work, and a reader who compares one finished deliverable per person before and after.
- Is It a Dealbreaker If Your Best Candidate Doesn't Use AI? Separate refusal from unfamiliarity, then check whether an assistant has reached the job's core task. Non-use only sinks a candidate where the work already runs on it.
- How Can You Tell If a Candidate Really Uses the AI Tools They List? Sort the six into categories and ask one question per category: what did it get wrong, how did you find out, what did you change. The thin answers are predictable.
- A Central AI Team Buys Demos; Embedded Capability Buys Work Centralize tooling and the verification standard, keep delivery in the functions that own the work product, and write the sunset condition before the first hire.
- Description Is Trainable; Discernment Is What You Hire For Specifying a task is a habit a new hire picks up in a fortnight. Noticing that a confident answer is wrong runs on domain knowledge onboarding cannot install.
- The Case for Still Hiring Juniors When AI Does Junior Work Entry-level hiring is a task-by-task call now. Keep the roles whose work still gets read and graded, cut the ones producing drafts nobody opens, and price the lag.
- A Candidate's Eyes Keep Moving Off-Camera: Cheating or Nerves? Off-camera eyes prove neither cheating nor nerves. Eye movement doesn't track deception, and acting on it carries a disability risk. Score the answer, not the gaze.
- How Do You Structure an AI-Heavy New Hire's First 90 Days? Stop reading the finished work. For three weeks make the framing note, one intermediate artifact and one named check the deliverable, then remove them and watch.
- First Round Collects the Claim, Final Round Watches the Work Round one collects claims and picks one worth testing. The last round produces what a conversation cannot: a slice of the real job, done with an AI assistant.
- What Follow-Up Questions Expose Whether Someone Understands the Answer They Gave? The follow-up that works asks for the alternative the candidate discarded. Then move a constraint and ask what changes, then ask what would prove the answer wrong.
- How Many People Do You Need If AI Takes Part of the Workload? Task exposure varies several-fold across occupations, so one productivity assumption misses high and low at once. Plan a range per occupational family instead.
- Should You Hire for AI Skills or Train the Team You Have? Train for the tool gap, hire for the verification gap. Last quarter's escaped errors tell you which gap your team has, and your field tells you how big it is.
- Hire a Person or Hand the Work to AI: How to Decide A salary against a subscription is the wrong comparison. Sort the tasks into producing and deciding, count the review hours automating creates, then open or hold the req.
- Collect the Scores Before the Debrief, Not During It Ratings submitted and locked before the meeting, the spread read out first, discussion only where scores diverged, and a decision rule written before the loop opened.
- How Do You Stop Hiring Managers Rejecting Candidates for 'Sounding Like AI'? A memo won't stop it. Require every rejection to name the claim that's wrong, the check run and what it returned, then read flag rates by req and by reviewer.
- How Many Interview Rounds a Role Actually Needs Each stage should be the only place one claim about the candidate gets evidence. Name the claim, delete the stage that cannot, and most loops collapse to three.
- What Interview Questions Show How a Candidate Works With AI? Ask about one recent task, not their philosophy: the first prompt, what got thrown out, what got checked. Twelve questions, plus the answer that should worry you.
- Twelve Interview Questions That Survive an AI-Prepped Candidate Twelve questions built around something checkable in the room: an artifact, a deletion, a claim that turned out wrong. Plus how far to take the follow-ups.
- Four Lines Worth Scoring on AI Use, and Two to Delete Four lines to put on the interview scorecard for AI use, each anchored to a quote or an artifact, and the two lines to take off the form entirely.
- Should You Hire the Junior Who's Fast With AI Over the Senior? Seniority doesn't decide this. Hire for who catches a confident wrong answer in your field, and let the cost of one undetected error pick the candidate.
- Where Juniors Get Their Reps When AI Takes the Grunt Work Pick two or three tasks that historically built judgment, make them narrated rather than banned, put them on a calendar, and give one senior the teaching.
- The People Best With AI Leave When Their Judgment Stops Mattering People good with AI leave when their objections stop changing what ships. Count the overrides, then give the check a named point, an owner, and a deadline rule.
- Four Levels of AI Skill, Written So Two Managers Agree Accepts, edits, checks, refuses. Each rung names something you could watch a person fail to do, which is the only property that makes a level usable for hiring.
- Four Questions for a Manager Whose Team Ships With AI Four questions about decisions: what the team automated, what the manager refused, how they handled over-reliance, and what the review caught last quarter.
- How Do You Get Managers Who Don't Use AI to Judge AI-Assisted Work? Skip the AI course. Give managers three questions and an answer key they write themselves, with strong and weak answers for engineering, marketing, finance and legal.
- A Manager Who Has Never Used AI Cannot Review AI Work Reviewing AI-assisted work takes recognition of three failure shapes, and that comes from hours in the tool. A course does not supply it. Here is the dose.
- The Minimum AI Literacy Is Three Judgments, Not a Curriculum What may be handed over, what has to be verified, and what data never goes into a prompt. Everything else, including how the technology works, is optional context.
- New Grads Who Never Worked Without AI: Training Gap or Hiring Mistake? A missing fundamental is trainable in weeks. Not noticing one was needed is a hiring signal. How to separate the two, and which fundamentals each field has to keep.
- Should You Care That New Hires Ask AI Before the Team? Asking AI first is not disloyalty. It matters only where your team's own context is the only check, which is exactly the context a new hire lacks.
- Say the AI Rule Out Loud in the First Two Minutes The opening script that gets a usable answer: name the rule, name the reason, promise the disclosure carries no penalty, then ask about choices rather than compliance.
- Give Each Panel Seat One AI Question Nobody Else Asks Four seats, four stages: what got handed to the model, how it was checked, what happened when it came back wrong, and what never gets delegated.
- Brilliant in the Screen, Flat Onsite: Which Candidate Is Real? Neither round is the real candidate. Trust the one whose conditions match the job, expect the format to move ratings, and ask what any assistance actually produced.
- Proctoring or an AI-Open Assessment: Which Stops Interview Cheating? An AI-open task removes the reason to hide the assistant. Proctoring runs an arms race against an overlay built to beat it, and its flags land unevenly.
- Promotion Criteria That Survive AI-Assisted Output Output volume rose for everyone at once, so shipped surface area no longer separates levels. Promote on the decisions instead, and make the packet prove them.
- Prompt Engineering Is Not the Skill Your Team Is Missing Three habits outlast any model release: specify before generating, source the claim that carries the decision, and test the output against something outside the chat.
- How Do You Redesign an Interview So AI Assistance Becomes Signal? Put a constraint the model can't know inside the task (your test suite, last quarter's churn, the client's real budget), then score what the candidate did about it.
- What Do You Hire For After AI Didn't Absorb the Work? The tasks that came back after the cut are the job description: exceptions, escalations, and checking confident output. Hire for those, not for the volume AI absorbed.
- What a Remote AI-Heavy Posting Has to Say About Review Remote removes watching someone work. AI removes the rough draft that used to get corrected. Say who reviews the work, against what, and what stays with the person.
- Build the AI Question From the Role's Own Tasks in an Afternoon A four-step derivation that turns the three tasks a role actually performs into two or three AI questions with a right answer, in any function.
- Anchor the Question to Their Work Instead of Rotating It Every Quarter Rotation breaks the one property that makes an interview predict anything. Publish the competencies, anchor each question to something only present in the room.
- Why Does Every Candidate Give the Same Polished STAR Answer? STAR is a fixed four-beat template, so an assistant fills it from your job posting in one prompt. Probe the discarded option, not the story.
- Send the Questions in Advance and Change What You Score Withholding no longer buys spontaneity now that preparation is universal. Send the questions, keep back the material, and score what advance notice cannot manufacture.
- STAR Survives If You Score the Result and the Discarded Option Three of STAR's four letters were always cheap to produce. Keep the frame, move the weight onto how the result was measured and what the candidate chose not to do.
- How Do You Run a Structured AI-Use Interview That Compares Candidates? Same questions in the same order, one rating scale, and behavioral anchors written before the first interview, plus why those anchors change from role to role.
- Do Structured Interviews Still Separate AI-Coached Candidates? Structure survives AI coaching; the shared question bank does not. Draw six to eight questions from decisions your team actually made, and ask what happened.
- The Tasks Worth Keeping Out of the Model Security lists answer a containment question. Three structural tests decide which recurring tasks a model may assist with and which it must never produce.
- Before You Open the Req, What Can Your Team Already Do With AI? Pull three recent deliverables per person and mark where AI touched each one, what got checked, and what got kept. The gap is usually verification, not output.
- What Separates Heavy AI Use From Good AI Use Usage dashboards count seats and prompts, not judgment. Four traces separate good AI work from a lot of it, and a ten-minute work review finds all four.
- How Do You Test Whether a Candidate Notices AI Errors? Plant one field-plausible error in the material, let the assistant repeat it, then grade four acts: whether it was named, when, what settled it, and what changed.
- What Do You Do When the Interviewee Isn't Who Did the Take-Home? Don't study the video and don't ask for ID. Extend the round for every finalist: a walkthrough of their own submission, on decisions nobody could rehearse.
- What Does Being Good at Using AI Look Like in an Interview? Framing before generating, evidence for the claim that matters, a test against something outside the chat. Ask for the verification move; it belongs to the field.
- Structure Is Four Things, and Same Questions Is Only One Job-analysed content, fixed questions, anchored rating scales, and independent scoring before discussion. Semi-structured names whichever of the four got dropped.
- Which Roles Actually Need AI Skills Right Now? Rank roles by how much of the occupation's task list a model already drafts, how often that work runs, and what a wrong answer costs. The output is three to six roles.
- Name the Human Who Checks It, or Nobody Did Accountability never moved: it belongs to whoever shipped the work. What has to be built is the named check, on the named claim, done by a named person.