Is an agentic interview the same as a chatbot interview?

No. A chatbot follows a decision tree of pre-written branches. An agentic interviewer has a goal (assess a specific competency), reasons about the candidate's actual answer, and decides its next move dynamically — closer to how a skilled human interviewer works than to a scripted bot.

Does the agent make the hiring decision?

No. The agent conducts the conversation and produces structured, evidence-backed scoring for a human to review. Final hiring decisions stay with the hiring team.

Can candidates game an agentic interview the way they might game a fixed script?

It's harder to. Because follow-ups are generated from the candidate's own answers rather than a fixed list, memorized talking points that don't hold up to a follow-up question tend to surface quickly.

What happens if the agent misreads an answer?

Every session produces a transcript and reasoning trail a human reviewer can audit, and scorecards are designed to be overridden — the agent narrows down where to look, it doesn't get the final say.

What Is an Agentic Interview?

Most "AI interviews" today are really just a form with a microphone: a fixed list of questions, played in order, with a speech-to-text layer bolted on. The candidate answers, the system moves to the next question, and a human sifts through the transcript afterward. It works, but it isn't fundamentally different from a paper questionnaire — it's just faster to administer.

An agentic interview is a different architecture. Instead of stepping through a script, an AI agent holds a goal (evaluate a specific skill or competency), listens to what the candidate actually says, reasons about whether that answer provides enough signal, and decides in real time whether to probe deeper, redirect, or move on. It behaves less like a survey and more like a competent interviewer who's read the job description and is actually listening.

What "agentic" actually means in an interview

"Agentic" gets used loosely, so it's worth being precise. An agent, in this context, is a system that:

Has an explicit objective (e.g., "assess this candidate's system design judgment")
Observes the state of the conversation so far
Decides its next action based on that state, not a pre-written index
Can take multiple different paths to the same objective depending on how the candidate responds

A scripted interview has none of this. It has a list. Question 4 always follows question 3, regardless of what was said. An agentic interview might ask question 4 to one candidate and skip straight to a harder follow-up for another, because the first candidate's answer already demonstrated depth that the second candidate's answer didn't.

How this differs from scripted AI interviews

The distinction shows up most clearly in three places:

Follow-up questions. A scripted system either has no follow-ups or has a small, pre-written set attached to each question. An agentic system generates its follow-up from the specific gap in the candidate's answer — it's reacting to content, not selecting from a menu.

Pacing. Scripted interviews take the same amount of time on every topic for every candidate. An agentic interview spends more time where the signal is ambiguous and moves faster through areas where the candidate has already demonstrated clear competence, which is closer to how a good human interviewer allocates attention.

Failure mode. When a scripted interview goes wrong, it usually produces a shallow transcript — technically complete, but low on signal. When an agentic interview goes wrong, it's usually because the agent misjudged which thread to pull on, which is a narrower and more auditable failure than "the script wasn't built for this candidate."

What the agent is actually doing behind the scenes

Under the hood, an agentic interview loop typically runs something like:

Load the role's evaluation criteria and the current conversation state
Listen to the candidate's answer (via ASR)
Reason about which criteria that answer does and doesn't address
Decide: ask a clarifying follow-up, move to the next criterion, or wrap the topic
Generate the next question or prompt in natural language
Score the exchange against the rubric and update the running evaluation

That loop repeats for the length of the interview. The candidate experiences it as a conversation. The hiring team gets a structured evaluation, tied to specific evidence from the transcript, at the end.

Why this matters for hiring teams

Better signal per minute. Time isn't wasted on questions the candidate has already answered well, and more time goes to the areas that are actually ambiguous.
Consistency without rigidity. Every candidate is evaluated against the same rubric, but the path to demonstrating competence isn't forced into an identical shape for everyone.
Fewer shallow transcripts. Because the agent probes vague or generic answers instead of accepting them at face value, reviewers spend less time trying to infer signal that was never actually captured.

Why this matters for candidates

A rigid script punishes candidates whose strongest answer doesn't happen to line up with the next scripted question. An agentic interview is built to follow the candidate's actual answer wherever it leads — which means a strong answer gets explored instead of skipped past, and a thin answer gets a fair chance to be clarified rather than silently counted against them.

It also tends to feel less like filling out a form. Candidates are responding to something that responded to them.

Where agentic still needs a human in the loop

None of this replaces human judgment — it front-loads it. The rubric the agent evaluates against, the criteria that matter for the role, and the final hiring call are all set and made by people. The agent's job is to run a consistent, adaptive first conversation and hand back structured, evidence-backed findings — not to be the last word on whether someone gets hired.

Getting started with agentic interviews

If you're evaluating whether to move from scripted AI screens to an agentic approach, start with one high-volume, well-understood role. Define the rubric clearly, run it alongside your existing process for a few weeks, and compare the transcripts side by side. The difference is usually obvious within the first handful of sessions: scripted transcripts read like a form; agentic transcripts read like an interview.