Where AI actually sits in the process
On a backend developer search this year, the funnel ran like this: 80 plus resumes screened, 20 taken through an AI interview stage, 10 shared with the client, 1 hired. Ten submittals for one hire, and a stage in the middle that most agencies do not have at all.
Most recruitment still runs CV screen straight to submittal. That is exactly why clients end up with candidates who read well on paper and interview badly. Adding a stage in between, an AI interview before a human ever reviews the shortlist, is what let half of the 20 who cleared the resume screen get filtered out before anyone's time was spent on them.
The developer search: what the AI interview stage filtered
That halving is the actual story. The AI interview did not replace judgement. It asked every one of those 20 people the same questions, in the same order, scored against the same rubric, with no fatigue and no drift over a long day of screening. What survived went to a recruiter, not straight to the client.
Ten of the twenty did not clear that stage. The point of publishing that number is not to show the AI being clever. It is to show that the filtering happened before a human's time got spent on it, and that a person still reviewed everyone who came through the other side.
What AI can judge, and what it cannot
An AI interview is good at consistency. Every candidate gets the same questions in the same order, scored against the same rubric, with no favouritism toward whoever interviewed last and no drop in attention on the fifteenth call of the day. That is a genuine advantage over a human doing back to back screens.
What it cannot do reliably is tell a fluent talker from someone who actually did the work. That distinction matters most in technical hiring, where the vocabulary sounds identical whether or not the person shipped anything.
The three-question method for the part AI cannot reach
Lokesh, who runs the search side of Animus Tech, is not from an engineering background and closes backend and frontend roles anyway. He took recruitment-tech courses to learn which skills actually matter per role, then built a verification method that does not require writing code himself. It checks three things, in order.
- Live project, or not. Was this shipped and used, or a college assignment, or a team product where this candidate's own piece is unclear? Someone who cannot draw that distinction himself is a flag on its own.
- What was actually owned. Not what the team delivered around them. What this person was personally accountable for.
- The problem statement. What problem was the project solving? Someone who owned the work states it in a sentence. Someone who was adjacent to it describes the technology instead.
That last question is the cheapest, highest-signal filter available, and it is very hard to fake convincingly across a real conversation. It is also something any hiring manager can ask themselves in thirty seconds, AI tooling or not.
The keyword problem AI alone does not solve
On an unrelated search, a candidate surfaced whose CV was, in Lokesh's words, not a resume but a page consisting of education, company name and internship name. No responsibilities listed, no achievements, nothing for a keyword match to grab onto. A pure keyword system, human or automated, would have skipped that CV entirely, regardless of who the person actually was.
An AI interview stage does not fix this by itself. It only gets a chance to evaluate someone once that person has already been pulled into the funnel. Sourcing that goes wider than keyword matching, and a recruiter willing to look past a thin CV, still has to happen upstream of any AI stage. Tooling narrows a pool that already exists. It does not go looking for the pool.
What we actually commit to
Our AI sources, screens and interviews at a scale no manual team matches. But nobody is rejected by software alone. A recruiter reviews every shortlist before it reaches you, because the questions that predict success on the job, like the problem-statement question above, are not the kind an algorithm asks well on its own.
That is the honest version of AI-powered: tooling narrows the field fast and consistently, and a person makes the calls that require judgement. If an agency tells you AI does the whole job, ask to see the funnel. If they cannot show you one, they probably do not have a real stage, just a label.
More on how we structure a search end to end is on how we work, and what we look for role by role is on our talent acquisition page. For software roles specifically, see IT and software recruitment in Pune.
Read next: How we work, and what we commit to ยท IT and software recruitment in Pune