Artificial intelligence is no longer just about breakthroughs in labs or pumping billions of dollars into data centres — it’s in our hospitals, courtrooms, classrooms, and on the battlefield. 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 just curious — this is where the signal cuts through the noise.
In this edition of the newsletter:
1 June 2026
Get the latest issue of Outlook Business
AI workers under squeeze
Cracking the Hinglish Vinglish puzzle
India’s AI stack missing key pieces
Going rogue for crypto
Humans in the Loop
AI workers under squeeze
It has been conclusively proven that the observable universe is expanding rapidly. But there is something that is probably expanding faster–the world of AI tools.
Every day, a new AI product hits the market and their makers shout from the rooftops about how the widget makes tasks — ranging from video editing and marketing to research and consulting — much easier to do.
What gets lost in the hullabaloo is the high cost of using these tools. The biggest companies in the world like Uber and Salesforce are complaining of how they are racking up eye-watering AI bills.
The biggest impact of these tools, however, has been on gig professionals – or ‘prosumers’ — using them. While clients are negotiating the price of work on the seeming ease of accomplishing a certain task. Professionals across sectors are complaining of misplaced pricing expectations from potential clients.
Gopi Balam, a Hyderabad-based video editor, has been working with different Indian and international clients, creating over 500 short and long-form videos every year for over half a decade. “Negotiations were always there, but the nature of negotiations have changed in the last one year,” Balam says.
With photo and video generating tools like Midjourney and Runway becoming household names, clients have started questioning the fees being charged. The launch of every new tool is followed by influencers making big claims of editing or research becoming as easy as typing a prompt. The reality is, however, costlier. What looks like a one-click output is often built through multiple paid attempts. Even though the tools have made the workflow more efficient, they come with a cost.
“AI gives output, but not final output. For example, if I use Runway or Pika, I generate small clips, 3–5 seconds. Then I have to regenerate many times to get correct motion, lighting and face,” says Balam. And every clip generation consumes credits or tokens.
He gave a rough breakdown of costs. While editing fees vary, an experienced editor typically charges at least Rs 4,000 for a one-minute video. Even if 30 seconds of the video is of AI generated clips stitched together, it would cost around Rs 1,000 in credits considering multiple iterations for every 4-6 second clip. And this is over and above the cost of editing software, skill and time.
Even professionals like designers and lawyers have started complaining.
“Costs have shifted more than reduced. Now we are paying for multiple tools like Midjourney, ChatGPT, and sometimes specific UI tools,” says Nayan Sharma, a Delhi-based marketing agency founder. Even for many small teams, AI has become another large overhead, not a cost-saving tool.
There is also little evidence that AI is consistently cheaper than human labour. A 2024 MIT study found that AI automation made economic sense in only about 23% of roles where vision is central to the work, with the remaining 77% still cheaper to be performed by humans.
More recently, Nvidia’s vice president of applied deep learning, Bryan Catanzaro, noted that for many teams, AI costs are already exceeding employee costs.
While monthly subscription plans take care of some of the needs, the usage limits often get exhausted with bulk usage. For an independent service provider or a small firm, the top-tier subscription is too expensive and the requirement of multiple tools increases the overall costs exponentially. Nayan points out that even traditional software providers like Adobe have added a layer of AI offerings and increased prices.
“Clients see output and think it is easy, so, undervaluation is happening,” says Saptarshi Ghosh, a Kolkata-based lawyer. “Many clients say they can generate basic agreements like NDA, MSA draft from ChatGPT, why is your fee so high?”
Ghosh laments that in sensitive sectors like law, it is not possible to completely rely on generic tools like ChatGPT and specialised tools like Harvey are not as pocket-friendly as people commonly think. In high-stakes work, the cost of getting it wrong still outweighs the cost of doing it right.
AI has made parts of the work faster. But it has not made it cheaper in the way many assume. For gig professionals, this has become a battle of perception. They are expected to deliver better output, faster, and at a lower price, even as their actual costs remain the same or, in some cases, rise.
From the Trenches
Inside AI’s cybersecurity nightmare
With the leading AI models gaining strength with every new update, the conversation amongst key stakeholders is moving from capability to security. While increased capability is leading to efficiency gains and even innovation in many cases, in the hands of bad-faith actors, the same capability is likely to become a disruptive force.
The solution is not to curtail usage, but to meticulously build guardrails around data and access, says Zoho’s AI Security Head, Sujatha S Iyer.
“Suppose you have an LLM with tool-calling facilities, a payroll agent. If you ask it queries, it will access payroll data and help draft an email. But what sort of data should the agent be able to access?” Iyer asks.
“If I ask the payroll system about my own salary or my basic pay, I should get an answer. But if I ask for the average salary of software engineers inside the company, that should not come through, because it means I am indirectly accessing everyone else’s data,” she adds.
To understand how this shift is playing out, it helps to look at how traditional cyber threats have evolved alongside these capabilities.
Phishing, for instance, has come a long way from its early days. “Phishing emails a decade back were poorly worded. They had bad CSS, bad HTML. But today’s phishing emails are picture perfect,” she says. The attacks have also become more targeted. What marketers call lead enrichment, attackers are now doing at scale, studying individuals and organisations before launching spear phishing campaigns. As a result, older detection methods based on grammar errors or suspicious domains are losing relevance.
The problem, Iyer explains, is not just the sophistication of attacks but the way enterprises defend themselves. “The biggest blind spot is that enterprises often look at things in silos,” she says. Endpoint security, browser security and internal monitoring operate independently, missing the broader pattern. Most attacks unfold across layers, where a phishing link can lead to malware and then data exfiltration. Without a unified view, organisations are reacting to fragments.
Zoho, which offers a wide suite of business software across CRM, finance and workplace tools, has been building systems that bring these layers together. The core idea is that security cannot function as isolated tools, it has to work as a coordinated system.
The risks become sharper for startups. “If you are a data startup or aggregator, your biggest risk is the data itself,” Iyer says. External threats like ransomware aim to lock or extract data at scale, while internal risks come from misuse of access. Traditional systems often detect these only after damage is done. AI-powered security is shifting the focus to early detection, anomaly tracking and real-time response.
At the same time, enterprises are deploying agentic AI systems that connect workflows, creating a new layer of vulnerability. “The key risk lies in the data access layer,” Iyer notes. If permissions are loosely defined, an AI agent can expose sensitive information. Employees also copy data into external tools and bring outputs back, creating what Iyer calls a “shadow AI” problem. Once data leaves the system, there is no visibility on its use. Integrating AI within enterprise tools, with defined permissions, reduces this risk.
Looking ahead, the pace of change itself is becoming a challenge. AI models are now capable of identifying vulnerabilities in code, something once limited to human experts. But the same capability can be used by attackers. “Every vulnerability you find should be treated as a zero-day,” Iyer says. What appears minor can be chained into a larger breach.
In that sense, cybersecurity in the AI era is no longer about fixing known problems. It is about assuming every gap can be exploited faster than before, and acting on it immediately.
Numbers Speak
India’s AI talent pipeline has a leakage
India is emerging as one of the largest producers of AI talent globally, but retaining that talent remains a challenge.
According to the 2026 AI Index Report by Stanford University, India has the second-highest number of AI authors and inventors at 50,460, behind the United States at 220,520. Germany follows closely with 48,520, while the United Kingdom and Canada form the next tier.
The absolute numbers suggest scale. India is clearly a major contributor to the global AI talent pool. But the picture shifts when adjusted for population. Countries like Switzerland and Singapore lead in per capita AI talent, while India does not feature among the top nations on that metric.
The bigger concern, however, lies in talent movement.
“Mobility is measured through net flow, which is the difference between the number of AI authors and inventors who move to or out of their respective countries,” the report notes.
By that measure, India recorded the largest net outflow of AI talent at -16.9 in 2025. In contrast, countries like the United States, Saudi Arabia and Denmark continue to attract more talent than they lose.
The contrast highlights a structural imbalance. India is producing talent at scale, but a meaningful share of that talent is finding opportunities elsewhere.
At a time when countries are racing to build sovereign AI capabilities, talent is as critical as compute and capital. The ability to retain researchers and engineers will shape where foundational innovation happens.
For now, India’s role remains that of a major supplier in the global AI ecosystem, even as the next phase of value creation increasingly depends on where that talent chooses to stay.
Words of Caution
Smarter AI, faster attacks
AI agents are getting smarter with every release, but they are also changing how cyberattacks are carried out.
India’s cyber security agency, CERT-In, has issued a high-severity advisory warning that newer frontier AI systems are significantly increasing attack capabilities. These models can identify vulnerabilities, generate exploits, and execute multi-stage attacks with minimal human input.
The shift comes from agentic AI models. Unlike earlier systems that responded to prompts, these models can plan tasks, use digital tools, and keep working until a goal is achieved.
In one recent instance, an AI system reportedly uncovered 271 previously unknown vulnerabilities in Mozilla Firefox, issues that had gone undetected for years.
For companies, the risk is clear. As AI becomes more autonomous, threats are becoming faster, scalable and harder to detect. Guardrails around access, permissions and deployment are no longer optional.
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)
