The Inference | AI and the Return of Old-Fashioned Hiring; India Inc Takes It Slow

Dear Reader,

Artificial intelligence is no longer just about breakthroughs in labs or billions poured into data centres – it's in our hospitals, courtrooms, classrooms and battlefields. At Outlook Business, we believe that India needs a sharp, nuanced and people-first lens on this transformation.

The Inference is our attempt to make sense of a world being rewritten by AI. In this newsletter, we bring you frontline narratives, boardroom insights and data you can trust. Whether you're an investor, founder, policymaker or simply curious – this is where the signal cuts through the noise.

In this edition of the newsletter:

• AI and the return of old-fashioned hiring

• The path beyond tokenmaxxing

• India Inc takes it slow

• AI agent goes all-out for master

Humans in the Loop

AI and the return of old-fashioned hiring

There are a few moments in life when even the calmest of people feel a bit uneasy, the well-prepared ones feel anxious and the confident ones start questioning themselves. Job interviews are one of them. Not surprisingly, many are now turning to AI tools to ace job assessment tests.

Nine months ago, Shaifali (name changed), a tech content creator was about to send out an offer letter when she spotted a ChatGPT window reflected in the candidate's glasses during a final video round. It set off an alarm bell in her head. Would this candidate be able to come up with original ideas?

Shaifali is not the only one getting cold feet while making a hire. Recruiters across the world are reporting the use of AI tools to ace job assessment exercises — and are being compelled to rethink the hiring process.

During the pandemic, at the height of remote work and a rapid upsurge of digitisation across industries and job functions, many processes that had traditionally depended on physical interaction moved online almost overnight. Recruitment was one of them. Online assessments, virtual interviews and remote onboarding became normal because companies had little choice, leaving the process vulnerable to the misuse of AI.

"During digital assessment, one can do multiple things. Even if you are under camera surveillance, you can still figure out multiple ways and means to get help from other channels," says Sanjaya Gupta, director at Bengaluru-based Spectrum Consultants.

The problem is no longer limited to looking up an answer on Google. Candidates can keep AI running on another device, use assistants that listen to questions during video interviews or generate code during online tests. Generative AI has made such assistance much faster and more precise.

Samriddh Singh, an IT professional in Noida who has recently been interviewing for new roles, says ChatGPT is now routinely used for preparation among his peers. But its use does not always stop there. He regularly hears of candidates keeping AI open in a second window during tests, or trading prompts that work better.

The response is beginning to look decidedly old-school. After years of pushing recruitment online, some companies are bringing parts of the hiring process back into offices, particularly for engineering and other critical roles where employers want to know whether the person behind an impressive assessment can actually do the job.

"Earlier, clearing the coding test was a big thing, now they want to check if you actually understand what you've written," Singh says.

Spectrum, a staffing firm, saw the same gap between test scores and actual understanding while handling assessments for a large client. Candidates who performed reasonably well in tests taken from home sometimes struggled with basic concepts during subsequent interviews. The company experimented by bringing candidates in one city to its office for supervised assessments. The pass rate did not change dramatically, but the candidates who cleared the supervised tests held up better in the interviews that followed – enough for the client to retain the physical format.

The shift, however, is not happening uniformly across the hiring process.  Pasupathi S, COO of HirePro, an AI-powered recruitment platform, says many companies are increasingly using a tiered approach: early screening may still happen remotely, but shortlisted candidates are then called in for supervised assessments or face-to-face interviews. For these roles, the physical interaction also helps employers judge motivation, communication and fit beyond what an online test can reveal.

At the same time, hiring platforms are trying to make digital assessments harder to game. Pasupathi says HirePro and its peers are using screen and audio monitoring, AI-based proctoring and other behavioural signals to detect suspicious assistance. In some cases, interviewers may even ask questions outside the knowledge normally expected of someone in that role; an unusually precise answer can become one of several clues that an external AI tool is being used.

Some assessments are also adapting to AI rather than trying to keep it out altogether. Candidates may be allowed to use an AI tool to generate code and are then tested on whether they can identify gaps, correct errors and make the output work. "It's like an open book," Pasupathi says.

The pandemic pushed hiring online; AI is now forcing parts of it back. As it becomes harder to tell where a candidate's ability ends and machine assistance begins, the pendulum is swinging back. For critical roles, the physical assessment and face-to-face interview are returning not as relics of an older system, but as safeguards against a new one.

From the Trenches

The path beyond tokenmaxxing

It has been nearly four years since ChatGPT made generative AI mainstream. While consumer-facing tools are widely discussed, the utility of using AI in enterprises is still being worked out. Companies across sectors have spent billions of dollars deploying AI across customer support, software development, internal operations and analytics. The question now is whether AI is delivering measurable business value.

"We've moved away from the innovation phase into the deployment phase," says Arun Balasubramanian, managing director, India and SAARC, at software analytics platform Dynatrace. "Enterprises are looking at how much they deploy, what outcomes they are getting, who is consuming it and whether it is delivering measurable business benefits."

That reframing is also changing how companies define AI return on investment. While token optimisation remains important, Arun says enterprises are no longer looking at it in isolation. "They're looking at it holistically. Infrastructure, operations, governance, talent and deployment all tie back to business outcomes," he says.

It is a significant departure from the early days of enterprise AI, when success was often measured by the number of pilots launched or models deployed. Today, organisations are increasingly tracking customer experience, employee productivity, operational efficiency and business KPIs instead of technology adoption alone.

The benchmarks also differ across industries. A digital-native company may measure AI through faster customer acquisition, retention and personalised experiences. Traditional enterprises, on the other hand, are often trying to reduce technology debt, improve cross-functional collaboration and modernise legacy systems. What remains common, however, is the need to demonstrate tangible business impact.

"You're monitoring token usage, queries, decision cycles, models and even why hallucinations are happening," Arun says. Continuous monitoring, he argues, is becoming essential to ensure AI systems remain reliable while delivering the expected business outcomes.

The growing focus on efficiency is also influencing model selection. Rather than defaulting to the largest frontier models, enterprises are becoming more disciplined about matching models to specific workloads. Arun describes this as "AI prudence", where organisations focus on deploying AI in the most cost-effective way while ensuring governance, accountability and measurable returns.

Numbers Speak

India Inc takes it slow

India is one of the world's biggest adopters of generative AI tools. But when it comes to business adoption, it still trails some of its developing-world peers.

India is ChatGPT's second-largest market, with more than 100 million weekly users. Yet, according to the World Bank's World Development Report 2026, the share of formal-sector Indian firms – registered businesses, as distinct from the informal economy – using AI stood at just under 25% in 2025. In comparison, adoption in Kenya and Nigeria reached around 40%, roughly matching levels in the United States.

The findings come from a World Bank survey of 4,205 firms across seven countries: India, Nigeria, Kenya, Mexico, Jordan, Thailand and the US. India accounted for the largest sample, with 1,355 firms surveyed.

The pace of diffusion itself is remarkable. "With each past wave of general-purpose technologies, the lag in adoption between high- and lower-income countries has been shrinking," the report says. 

There is, however, an important caveat. The headline numbers capture whether firms use AI, but not how intensively they use it or whether that use improves business capabilities.

The World Bank also found that the gap with advanced economies widens as AI applications become more sophisticated. For India, the test is twofold: catch up with developing-world peers on adoption, and deepen AI integration to deliver measurable productivity gains.

Words of Caution

AI agent goes above and beyond for master

An AI agent was asked to book a gym class. It ended up exploiting the booking system and removing another person from the list.

In the first known Australian case of its kind, Andrew used an AI agent to secure a place in a coveted morning class, the country's public broadcaster ABC News reported. The agent discovered that the software lacked proper authorisation checks, booked classes beyond the permitted window and cancelled another person's reservation without being instructed to do so. It then reported that it could not restore the booking.

The incident highlights a growing risk with agentic AI. These systems do not merely answer questions; they can plan and act independently. When given access to poorly secured digital tools, even a routine task can produce unintended and potentially harmful consequences.

Best of our AI coverage

• Why India Doesn’t Have AI Start-ups in Growth Stage (Read) 

• Narayana Murthy's Message to India's Youth: Don't Settle for Being the World's Back Office (Read) 

• Bhavish Aggarwal's Speed is Breaking India's Only AI Unicorn (Read)  

• India’s Healthcare AI Start-Ups Grapple with a Broken Data Ecosystem (Read) 

• AI Start-Ups Ride a Wave of ‘Curiosity Revenue’, VCs Rethink What It’s Worth (Read) 

• What Exactly Is an ‘AI Start-Up’ — and Does India Have 5,000 of Them? (Read)

Podcasts for you