Why “can they do the job” or assuming “IIM-A + IIT-B with highest GPA can do the job” is the wrong now – and “can you do the job with AI” is the only one that matters
A wake-up call to every Head of Talent Acquisition still scoring candidates on skills a language model already has.
Let us start with a confession.
These days EdTech sells lots of certificates and every BFSI candidate you interviewed would have done a valuation certificate course with jazzy presentation skills learnt. Ask them a simple question on whether fair value per share should be computed after ESOP dilution or prior to ESOP dilution and no candidate would be able to give a complete answer. If interviewer knew the right answer (high chances the interviewer also does not know the answer) then all of them would be marked down. Meanwhile, any of these candidates could have got this done out of Claude.ai in fifteen minutes flat, if they knew exactly what to ask for, how to check Claude’s assumptions, and knew basics of fair valuation well (without being brain washed by sometimes wrong standard market practices on valuation followed by the industry and the government agencies in general).
Let us take an example of a start up with massive ESOPs (much higher than issued capital) and too with a deep discount exercise price vs listed companies issuing small amount of ESOPs at closer to market price – should the valuer adopt different methodology on potential dilution from ESOPs? Should the valuer adopt different fair valuation methodology or risk based assumptions based on who requests fair value – e.g. management for disclosures, investor to evaluate investment, Income Tax department to focus on tax evasion, NCLT or SEBI for takeovers and mergers or RBI for foreign exchange management compliances.
In any case you hired a candidate without testing their prompting skills (the product teams were pushing HR for a fast close). It just happened she had highest IQ levels – the one who can recite WACC formulas on a whiteboard. Thinks LLMs would make her dumb and needs to stay away from it. Six months in, she still writes prompts like Google search queries and treats every AI output as gospel and if pushed to get output from LLMs/ She would be draining tokens from poor prompting skills until she learns how and what to prompt, understands what output can be trusted from the LLMs.
The Interview Today Is Measuring the Wrong Organism
Every recruitment process in financial services is still built on a single, quietly false assumption: that the candidate’s own unaided knowledge is the thing being hired.
It is not. Not anymore.
The unit of productivity at work today is not the human. It is the human plus the AI they are operating. A brilliant analyst who can’t get useful output from an LLM is now less productive than a mediocre analyst who can. That sentence would have sounded absurd three years ago. Today it is simply an observation about how work actually gets done.
Imagine a coder without prompt engineering skills today!!
And yet walk into any assessment centre, any case interview, any campus hiring process, and the entire evaluation still happens in an AI-free vacuum – as if the candidate will show up to their desk on day one and be handed a locked laptop with the internet disabled.
You are evaluating a version of the employee who will never exist.
Two Candidates, One Score, Completely Different Outcomes
Here is the diagnostic that should be part of every interview from this point forward – because it is the diagnostic that actually predicts performance.
Sit two candidates down with the same AI tool and the same messy, ambiguous problem. Not a textbook case. A real one – an incomplete dataset, a contradictory client brief, a deadline that doesn’t allow for a clean answer.
Candidate A treats the AI like a search bar. One-line prompts. Accepts the first output. Cannot tell you why the model’s risk assumption is wrong, because they never learned to interrogate it in the first place.
Candidate B treats the AI like a very fast, very literal junior analyst. They scaffold the problem before prompting. They ask the model to show its reasoning. They catch the hallucinated data point on the second pass. They iterate three times in the time it takes Candidate A to get one usable draft – and their third draft is investment-committee-ready.
On a traditional resume, these two candidates might look identical. Same degree, same GPA, same case-competition trophy on the shelf. On the only test that predicts what they’ll actually produce at your desk, they are not in the same league.
This is the gap your hiring process is currently blind to. It is also the single largest predictor of who will be productive in an AI-augmented workplace – and almost nobody is measuring it.
Prompting Is Not a Soft Skill. It Is ‘the Skill’.
There is a reflexive dismissal that happens whenever “prompting” comes up in a hiring conversation – as if it’s a trick, a shortcut, something adjacent to the real work rather than the work itself.
That dismissal is going to be very expensive for the firms who hold onto it.
Prompting efficiency is not about knowing clever phrasing. It is about problem decomposition – breaking an ambiguous business question into the sequence of smaller, checkable steps a model can actually execute well. It is about specification – knowing exactly what output format, what assumptions, what constraints matter before you ask. It is about verification – knowing which parts of an AI’s answer to trust immediately, which to sanity-check, and which to throw out entirely.
Strip away the word “prompting” and look at what it actually requires: structured thinking, domain judgment, and the discipline to verify rather than accept. That is not a soft skill sitting next to technical competence. In an AI-augmented workflow, it is the technical competence – the layer that converts raw model capability into usable, trustworthy work product.
A candidate who prompts well is not “good with AI tools.” They are demonstrating the exact reasoning skill your job actually requires – just applied through a new interface.
What Your Job Readiness Score Should Actually Be Measuring
This is precisely why we built the Job Readiness Score and the Prompt War Game into the Zetheta WorkBridge platform the way we did – not as a gimmick, but because we saw this gap forming before most recruitment functions had a name for it.
A candidate’s project score on Zetheta doesn’t just measure whether the output was correct. It measures:
How efficiently they used AI to get there – fewer, sharper prompts producing a usable result, versus dozens of vague ones producing noise.
Whether they verified what the model gave them – catching an incorrect assumption, a fabricated citation, a miscalculated ratio – instead of shipping the AI’s first draft as their own.
How they integrated AI output into actual judgment – using the model to accelerate the boring 80%, so their own thinking goes into the 20% that actually requires it.
By the time a candidate reaches your interview through this pipeline, you are not evaluating whether they know finance. You already know they can produce enterprise-grade financial work under real time pressure, with AI as their working environment – because that is the environment they were tested in.
That is a fundamentally different signal than a resume. A resume tells you what someone studied. This tells you how they actually work.
The Evidence Is Already Sitting in Your Own Office
You do not need to take our word for this. Run the audit inside your own firm.
Pull your last twelve months of performance reviews. Cross-reference them, quietly, against who on your team is fastest with AI tools – not who has the fanciest ChatGPT Plus subscription, but who produces materially better output, materially faster, because they know how to work the tool and which tool to select for which problem.
The correlation will not be subtle. Your highest performers are very likely not the people with the most credentialed backgrounds. They are the people who figured out, on their own, how to turn a language model into a force multiplier – while everyone else was still using it as a fancier spell-checker.
Now ask the uncomfortable follow-up: does your hiring process select for that trait at all? For almost every institution, the honest answer is no. You are optimizing your funnel for a skill that no longer predicts performance, while the skill that does predict performance goes completely unmeasured until the person is already six months into the job.
This Is the Same Shift We Mapped for the Institution Itself
We have just published our strategic blueprint on this exact inflection point at the institutional level – The Invisible Finstitution. It argues that the financial institutions winning the next decade won’t be the ones with the best individual tools. They’ll be the ones that rebuilt their entire operating architecture around AI agents working with human oversight, rather than humans working with AI as an occasional favour.
Recruitment is where that shift starts, and where most institutions are getting it backwards.
You cannot build an agentic, AI-augmented institution by hiring people the way you hired them in 2015 and hoping they’ll “pick up AI on the job.” The employees who will thrive inside the institution the book describes are exactly the candidates your current interview process is failing to identify – because you’re still asking them to prove they can do the job without the tool they’ll be using for every single task, every single day, for the rest of their career.
The bank of tomorrow runs on human-plus-agent teams. So should your hiring funnel.
The Five-Year Gap That’s Already Opening
Here is the forecast, and it is not a subtle one.
Firms that redesign their hiring bar around human-plus-AI productivity in 2026 will spend the next five years compounding an advantage that gets harder and harder to close. Their analysts will be faster on day one. Their onboarding will be measured in weeks, not the industry-standard 18 months, because the skill gap they’re filling is domain context – not “how do I even use this tool.”
Firms that keep interviewing for 2015 competence will keep hiring 2015 employees. Competent ones. Credentialed ones. And measurably slower than the candidate down the street who never got past your resume screen because their college wasn’t on your target list – but who could out-produce your new hire by lunchtime on day one.
The talent gap in financial services was never really about a shortage of smart graduates. It has always been about firms measuring the wrong thing. The AI era didn’t create that problem. It just made it impossible to ignore any longer.
The Question for Your Next Interview Panel
Stop asking candidates to prove they can do the job alone in a room with no tools, because that room does not exist at your firm.
Start asking: how well can this person think with AI as a partner – decompose the ambiguous problem, prompt with precision, and catch the model when it’s wrong?
That is not a nice-to-have layered on top of “real” competence anymore. It is the competence.
The candidates who can do this are already out there, already building portfolios that prove it, on platforms designed to measure exactly this. The only question is whether your hiring process is built to recognise them – or whether you’re still filtering them out before they ever get in the room.