AI in hiring is usually introduced through a single question: can an algorithm decide who a company should employ? The more practical application of the technology sits several steps earlier.
A company posts a role. Fifty, a hundred or three hundred applications arrive. Someone has to review the candidates, schedule the initial calls, spend fifteen or thirty minutes with each of them, take notes, and only then decide who should speak to the hiring manager. It is a repetitive process that consumes expensive human time — and that is exactly what makes it a good candidate for automation.
The problem isn't the interview. It's the first interview.
Take a role that attracts a hundred applications, thirty of which reach the first round. If the initial conversation runs twenty minutes, the calls alone cost ten hours of work. Add scheduling, moved appointments, preparation and notes, and a single opening can easily absorb several working days.
And the goal of those calls is usually not to reach the final decision. The goal is to find the candidates worth having a longer conversation with.
The real question for AI is not who should we hire. It is which part of the process we can automate so that people spend their time where their judgement is worth the most.
The AI video interview as a screening layer
Job Bolt, the platform I built, approaches this as an AI video interview for initial candidate screening. The employer enters the job description and defines the questions. The candidate receives a link and takes the first video interview at a time that suits them. The AI runs a structured conversation and can ask follow-up questions based on the candidate's answers. The employer then receives the video interview, a transcript, a summary, an analysis of the answers, competency scoring and a comparison across candidates.
After: application → AI video interview → analysis and comparison → human review → conversation with the selected candidates
The recruiter or hiring manager no longer has to be present in every first conversation. The AI automates the repetitive part of the process, while the decision about who moves forward stays with the hiring team.
Why video still matters
If you automate screening on the CV alone, the candidate becomes a set of structured data. But a CV is not an interview. A candidate may not have a perfectly written CV and still explain very well how they solved a specific problem, why they made a particular decision, or how they think in a situation relevant to the job.
That is why Job Bolt keeps the complete video interview and transcript alongside the AI analysis. The hiring team can see the result of the analysis, then open the candidate's original answer and reach its own conclusion. The AI becomes a layer for organising and processing information, not a replacement for human judgement.
What was hardest to build
The idea was simple: if AI can hold a natural conversation, why not automate the repetitive first round of interviews? The implementation was considerably more complex.
The biggest challenge was not building an AI that asks questions. The problem was building a system that knows how to conduct an interview — to listen to an answer, understand the context, ask a meaningful follow-up question, and at the same time keep enough structure that we can later compare dozens of candidates consistently.
That required several connected layers: a real-time AI conversation, structured evaluation against competencies, evidence-based scoring, and a system for comparing candidates. In parallel, the video pipeline has to record both the candidate and the AI interviewer, store the interview securely while the conversation is running, and produce a complete recording for the hiring team at the end.
Consistency of the AI evaluation was a challenge of its own. A generative model can interpret the same answer differently, so a significant part of development went into structuring the analysis and tying scores to specific evidence from the conversation. That is where I changed my mind most about what it means to build an AI product.
The model is only one part of the system. The much larger job is making everything around it reliable enough to be used in a real business process.
A structured interview becomes structured data
When candidates answer the same core set of questions, the interview also becomes a structured source of data. The answers can be analysed against the competencies relevant to that specific role, and candidates can be compared on the same basis. A hiring manager can see why a candidate received a particular score and then check the original answer. That brings a consistency which is hard to achieve when different people run different conversations and take different kinds of notes.
You don't have to change your whole hiring system
A common mistake in business automation is trying to make a new AI tool replace the entire existing system. For this problem, that isn't necessary. Job Bolt was built as a layer for initial screening, not as a replacement for an Applicant Tracking System. A company can carry on using LinkedIn, its existing ATS, a job board or its own careers page. The candidate is simply directed to the AI interview, and the results are used before the next human round.
That is often the simplest way to introduce AI into a company: don't change the whole infrastructure, automate one clearly defined point in an existing process.
How to tell whether automation makes sense
Before implementing an AI system, it is enough to measure a few things. How many candidates enter the first round? How long does the average conversation take? How much time goes on scheduling and administration? How many good candidates never got the chance to reach a first conversation? How many people take part in the process?
If a company holds five initial conversations a month, it may not have a problem worth automating. If it holds a hundred, the arithmetic looks completely different.
That is a more useful way to think about AI than asking "where can we put AI?" First find an expensive, repetitive process. Measure it. Only then decide whether the technology can shorten it meaningfully.
From one AI workflow to connected systems
The AI video interview is only one point in a wider hiring workflow. The data produced during the process can be connected further to an ATS, internal databases, analytics, onboarding processes or reporting systems.
This is where the problem approaches the broader idea of AI infrastructure that MeteorIT develops through Integral: business systems stop being isolated islands of data and become sources that AI can search, analyse and use inside controlled workflows. In that architecture, the interview is not the end of automation — it is one of its sources.
The principle stays simple.
Don't automate because AI exists. Automate where there is a measurable cost the technology can remove.
