Airtel Priority Postpaid exposes blind spot in net neutrality framework

by Hansraj Agrawal | Jun 5, 2026 | Technology

For nearly a decade, India’s net neutrality framework has largely focused on one concern: preventing telecom operators from favouring certain websites, apps or online services over others. The debate that shaped India’s internet policy revolved around questions such as whether telecom operators could prioritise one app over another, slow down specific services, or offer differential pricing linked to content.

Airtel’s new Priority Postpaid plans raise a different question altogether: can telecom operators offer a better network experience to one group of subscribers if they are willing to pay for it?

The distinction may seem subtle, but it sits at the heart of the debate surrounding Airtel’s use of 5G network slicing technology. It also helps explain why regulators, telecom operators and policy experts appear to be viewing the issue differently from earlier net neutrality controversies.

What Airtel has actually launched

Bharti Airtel launched its Priority Postpaid plans on May 19. The company says the plans use 5G network slicing technology to offer a more dependable network experience during periods of congestion.

The promise is not necessarily higher internet speeds. Rather, Airtel says customers on these plans will receive greater consistency and reliability even when the network is under stress.

According to the company, the service operates through a dedicated network slice while remaining completely content-neutral.

In submissions to the government, Airtel argued that the implementation involves no blocking, throttling, preferential treatment of applications, or zero-rating. The company also maintained that the service does not degrade the experience of prepaid customers, who account for around 92 per cent of its customer base and contribute roughly 88 per cent of its revenue.

Airtel further stated that its overall 5G capacity utilisation during busy hours is around 38 per cent. Within that, postpaid traffic accounts for only about 4 per cent and may rise to roughly 6 per cent after the priority slice is introduced.

Why this debate is different from earlier net neutrality battles

Much of the discussion around Airtel’s plans has focused on whether they violate net neutrality. The more interesting question is why the answer appears to be different from earlier telecom controversies.

India’s net neutrality debates in the 2010s were largely centred on content discrimination. Regulators were concerned about whether telecom operators could favour certain websites, apps or online services over others.

The concern was whether a telecom operator could slow down one service, prioritise another, or offer differential treatment linked to the content being accessed.

Airtel’s Priority Postpaid plans do not neatly fit into that framework.

The company is not prioritising Netflix over YouTube, or WhatsApp over Telegram. Instead, the differentiation is occurring at the subscriber level.

The debate is no longer about whether different online services are treated differently. It is about whether different categories of users can receive different levels of network experience.

That shift is what makes the Airtel controversy different from previous net neutrality disputes.

Understanding 5G network slicing

To understand why this distinction exists, it is necessary to understand what network slicing actually is.

Network slicing allows telecom operators to create multiple virtual networks on top of the same physical network infrastructure.

A simple way to think about it is through a highway analogy.

Traditionally, all users travelled on the same highway. Traffic conditions could vary, but everyone shared the same road. With network slicing, telecom operators can create multiple virtual lanes on the same highway, each designed for a specific purpose.

One lane may be optimised for reliability, another for low latency, and another for specialised use cases.

According to industry reports, network slicing is a standard capability built into 5G architecture and is already being deployed in markets such as Singapore, the United States, the United Kingdom and Malaysia.

The technology itself is not the source of controversy. The debate begins when that capability is used to create differentiated consumer experiences.

Why TRAI may not view it as a net neutrality violation

Recent reports suggest that India’s telecom regulator is leaning towards the view that Airtel’s plans do not violate existing net neutrality norms.

According to a Business Standard report, TRAI’s examination found that the plans did not appear to degrade the quality of service or experience of prepaid users.

People familiar with the deliberations told the publication that net neutrality concerns generally arise when there is discrimination based on content, which does not appear to be the case here.

Another person cited by the publication said that the rules specify discrimination between customers of the same class, and that no such discrepancy had been identified so far.

The regulator is reportedly continuing to monitor implementation, but the emerging view appears to be that Airtel’s service remains within the boundaries of the current framework.

What India’s net neutrality rules actually say

Understanding the regulatory debate requires looking at the framework itself.

India’s net neutrality regime rests primarily on two foundations.

The first is TRAI’s 2016 Prohibition of Discriminatory Tariffs for Data Services Regulations, which prohibit service providers from charging discriminatory tariffs based on content.

The second is the Department of Telecommunications’ 2018 net neutrality framework, which was subsequently operationalised through amendments to the Unified Licence and ISP licensing framework.

Clause 2.3 of the Unified Licence requires internet access services to follow a principle that restricts discrimination, restriction or interference in the treatment of content. The provision addresses practices such as blocking, degrading, slowing down, or granting preferential treatment to specific content.

The recurring theme across these provisions is content.

The framework was designed to prevent telecom operators from discriminating based on what users access online.

That focus made sense when the biggest concerns involved issues such as zero-rating, content prioritisation and differential treatment of internet services.

The 5G question the framework never explicitly addressed

This is where Airtel’s Priority Postpaid plans have sparked a broader discussion.

The existing framework focuses extensively on content-based discrimination. It says comparatively little about subscriber-tier differentiation that remains content-neutral.

That does not automatically mean Airtel’s implementation violates the framework. In fact, Airtel’s defence rests precisely on the argument that all content, applications and online services continue to be treated equally.

However, network slicing introduces a different dimension to the debate.

The Airtel debate asks whether telecom operators can offer differentiated network experiences to different categories of subscribers while remaining content-neutral.

The distinction may appear technical, but it is central to understanding why regulators, telecom operators and policy observers are debating the issue.

Why the telecom industry is backing network slicing

Airtel’s defence of Priority Postpaid is not surprising.

More notable is the support that the broader concept of network slicing has received from across the telecom ecosystem.

In a letter to the Department of Telecommunications, Reliance Jio described network slicing-based service deployments as a legitimate exercise of 5G capabilities and argued that the existing regulatory framework permits such offerings.

Jio also stated that network slicing is a standardised 5G architectural capability that serves diverse connectivity requirements.

Industry body Cellular Operators Association of India (COAI) has similarly argued that slicing does not violate current net neutrality principles because it remains content-agnostic.

Former BSNL chairman Anupam Shrivastava also told ET Telecom that 5G slicing complies with current net neutrality frameworks because it treats applications equally and uses advanced spectrum optimisation.

The broad support is significant because it suggests the discussion extends beyond a single Airtel product.

A significant part of the telecom ecosystem appears to view network slicing as a legitimate 5G capability with long-term commercial potential.

The search for a 5G business model

The telecom industry’s support for network slicing is also linked to a larger commercial challenge.

Historically, telecom operators monetised voice calls, SMS services and data packs. While 5G has brought significant technological improvements, generating new revenue streams from those capabilities has been a persistent challenge.

Network slicing offers a different possibility.

Instead of simply selling larger data packs, operators could potentially offer differentiated levels of network reliability, latency, consistency or congestion protection.

This is one reason Airtel has consistently advocated network slicing as a major monetisation opportunity.

According to Business Standard, the company has described slicing as the only proven large-scale monetisation model for 5G and a foundational capability for future 6G networks.

Viewed through that lens, Priority Postpaid is not simply a new postpaid plan. It may also represent an early attempt to commercialise differentiated network performance as a consumer-facing product.

Are operators monetising congestion?

During a discussion hosted by digital news platform The Federal, telecom analyst Parag Kar argued that regulators must ensure premium users do not receive benefits at the expense of ordinary users.

He also highlighted an important distinction between wireless and fixed broadband networks. Unlike fibre networks, wireless systems operate within finite spectrum constraints, making capacity allocation a more sensitive issue.

Kar raised a broader concern: if network congestion becomes a persistent feature of mobile connectivity, operators may eventually be incentivised to monetise that congestion rather than eliminate it.

Others on the panel argued that the focus should be on ensuring a minimum quality standard for all users.

Tech policy researcher Pranesh Prakash argued that regulators should concentrate on maintaining and enforcing baseline quality-of-service standards rather than prohibiting all forms of service differentiation.

The discussion ultimately points to a broader policy question: if premium services are permitted, what safeguards should exist to ensure that ordinary users are not adversely affected?

What this could mean for consumers

The significance of Airtel’s launch extends beyond a single postpaid plan.

If regulators ultimately conclude that network slicing-based differentiation is fully compliant with existing rules, similar offerings could emerge across the industry.

That does not automatically mean ordinary users will receive worse service.

Airtel has maintained that its implementation does not degrade the experience of prepaid customers, and TRAI’s reported preliminary assessment appears to be moving in the same direction.

However, the emergence of premium network tiers would raise broader questions about how network quality should be priced, monitored and regulated in the 5G era.

For years, India’s net neutrality debate focused on whether telecom operators should be allowed to treat internet traffic differently.

Airtel’s Priority Postpaid plans have introduced a different question: should telecom operators be allowed to offer different levels of network experience to different categories of users?

That question may shape the next phase of India’s telecom policy debate long after the current controversy fades.

Source link

Ads to algos: Are IG, Snapchat and YT becoming new digital shopping malls?

by Hansraj Agrawal | Jun 5, 2026 | Technology

Not long ago, product discovery largely began with search engines, ecommerce platforms, television advertisements or banner ads. Today, it increasingly starts with a scroll. A creator demonstrating a skincare product, a short video reviewing a smartphone, or an influencer unboxing a new gadget can shape consumer choices in ways traditional advertising often could not.

What began as a platform for social interaction is rapidly evolving into a marketplace for discovery, influence and transactions.

A Meta-Ipsos study, which surveyed more than 4,000 respondents across 23 cities spanning metros, Tier-2, Tier-3 and rural India, found that 97 per cent of users watch videos on social platforms daily. More significantly, the study highlighted the growing role of short-form video in the consumer purchase journey.

According to the findings, Reels influences 81 per cent of product discovery, 66 per cent of product consideration and 47 per cent of purchase decisions among surveyed users.

Similarly, YouTube Shorts now attracts more than 650 million monthly logged-in viewers globally. At the same time, YouTube reaches more than 75 million people in India, underscoring the growing role of video in how consumers discover content, creators and brands.

The shift is also reflected in advertiser spending patterns. According to Snapchat’s 2026 report, the number of advertisers on the platform in India has grown tenfold over the past two years, while the number of advertisers spending across all four quarters has tripled. Snapchat attributed the growth to rising adoption of immersive advertising formats, the emergence of Chat as a high-attention advertising surface, and investments in AI-powered targeting capabilities.

For businesses, these numbers point to a larger transformation. Social media is no longer merely a channel for marketing products; it is increasingly becoming the place where consumers discover, evaluate and decide what to buy. In effect, the scroll is turning into the storefront.

That raises a broader business question: if discovery, influence and purchasing decisions are increasingly happening within social platforms, are Instagram, Snapchat, and YouTube evolving into the digital shopping malls of the modern internet?

From ads to algorithms: The rise of discovery commerce

The growing influence of social media on purchasing decisions reflects a broader shift in how consumers discover products online.

For decades, the path to purchase followed a relatively predictable sequence. Brands invested in advertising, consumers encountered those advertisements through television, print, search engines or websites, and purchase decisions followed later. Discovery and commerce were often separate events, taking place across different platforms and over extended periods.

Short-form video is changing that dynamic.

Consumers no longer need to visit a marketplace or actively search for a product to discover it. Instead, recommendations increasingly emerge within social media feeds, where algorithms surface content based on user interests and engagement patterns.

In many cases, product discovery occurs while users are consuming content rather than actively shopping.

This phenomenon is often described as discovery commerce — a model in which content, creators and platform algorithms play a central role in introducing products to potential buyers.

According to the Social Commerce Playbook published by Economy Insights in September 2025, recommendation algorithms are increasingly matching products with highly specific consumer interests, enabling brands to reach audiences at moments of relevance rather than intent.

As a result, the traditional marketing funnel is becoming more compressed. Content drives discovery, discovery influences consideration, and consideration can quickly translate into a purchase decision, often within the same platform ecosystem. The funnel has not disappeared; it has increasingly moved inside the feed.

The trend is particularly significant in India because of the scale and spread of video consumption. According to the Meta-Ipsos study, daily video engagement has reached 98 per cent in urban India and 94 per cent in rural India, suggesting that short-form video consumption is now nearly universal across geographies.

The report notes that India’s video-first transformation is being driven not only by metropolitan markets but also by users across smaller cities and rural regions.

The evolution extends beyond content feeds. Snapchat has expanded Chat as an advertising surface, stating that Sponsored Snaps, which appear within active conversations, generate 2.5 times higher brand awareness than traditional in-feed formats.

The move reflects a broader industry trend: platforms are embedding discovery and commerce opportunities across multiple user touchpoints rather than limiting them to conventional advertising placements.

For brands, the opportunity for product discovery is no longer confined to consumers in large urban centres. As social commerce expands beyond metros, platforms are becoming an increasingly important gateway to the country’s next wave of digital consumers.

Why businesses are paying attention

Marketers have always followed attention. Right now, attention lives in short-form video, and the numbers are compelling enough for marketing budgets to follow.

According to the Meta-Ipsos study, 84 per cent of Gen Z consumers discover new products and brands through social platforms. Among rural users, that figure stands at 73 per cent.

These are not passive audiences; they are active discovery surfaces.

The same study found that Reels delivers nearly 60 per cent higher creator engagement than other surveyed short-form video platforms, making it a primary surface for creator-led storytelling and brand discovery.

The convergence is not limited to one platform.

According to Snapchat’s 2026 report, with more than 250 million users in India, Snapchat sits at the centre of how Gen Z discovers, communicates and engages with brands, reinforcing that the shift towards social-led discovery is a platform-wide phenomenon rather than a single-platform story.

The report also noted that while Gen Z pays up to 34 per cent less attention to ads on conventional social platforms compared with Millennials, immersive and creator-native formats continue to drive strong engagement and measurable business outcomes.

For established brands, this represents a shift in how media budgets are allocated.

For D2C startups and smaller businesses, however, it is something more significant.

A brand that could not afford television or print advertising can now reach millions through a single well-placed creator partnership or a piece of content that resonates with the algorithm.

The economics of customer acquisition are changing.

Creator-led content, when executed well, lowers the cost of reaching relevant consumers. It replaces the broad reach of traditional paid media with something narrower and often more trusted: a recommendation from someone the viewer already follows.

According to Billo’s 2026 social commerce trends analysis, social platforms have become the primary entry point for product discovery, with short-form video, user-generated content and creator-led content shaping purchase decisions long before checkout.

The report noted that entertainment and live shopping are increasingly collapsing the gap between viewing and buying by embedding commerce directly into content.

The winners of the shift

Several groups stand to benefit from this structural change.

Creators

Creators are no longer just content producers. They are trusted commerce intermediaries.

Their audiences follow them not because of polished production values but because of perceived authenticity. When a creator recommends a product, it often feels less like advertising and more like advice.

This trust is commercially valuable, and platforms are building infrastructure around it through affiliate programmes and revenue-sharing models.

Direct-to-consumer brands

D2C brands may be among the biggest winners.

Short-form video allows brands to tell their story, demonstrate products and generate demand without the traditional costs associated with retail shelf space, print advertising or large media budgets.

The platforms provide reach, creators provide credibility, and algorithms provide targeting.

For emerging brands, this combination can be powerful.

Social platforms

Social platforms themselves are transforming their business models.

What began as advertising-led businesses is increasingly becoming commerce infrastructure.

The more of the purchase journey that happens within a platform — discovery, consideration and transaction — the more valuable that platform becomes to brands.

The race to build native shopping experiences, affiliate systems and creator commerce tools reflects this shift.

Instagram has expanded creator affiliate programmes and shopping features, while YouTube has integrated shopping tools that allow creators to tag products directly within videos.

Rather than directing users to external websites, platforms are increasingly trying to keep discovery, engagement and transactions within their own ecosystems.

Influencer marketing agencies

Influencer marketing agencies are evolving from talent managers into performance marketers.

The shift from impressions to conversions is changing how these businesses operate.

The question is no longer, “How many people saw this?” but “How many people bought something because of it?”

The Meta-Ipsos study notes that creators are no longer simply content producers; they are trusted discovery engines actively shaping culture, trends and brand choices across Gen Z, women, rural India and premium audiences.

The agency sitting between a brand and that creator network is increasingly sitting between a brand and its next customer.

As Saugato Bhowmik, director of CPG, D2C and automotive at Meta India, put it: “Creators, culture and commerce are converging on Reels in ways we haven’t seen before. For brands, this isn’t just a content play — it’s an always-on content-to-commerce play.”

The shifting role of ecommerce platforms

The shift also has implications for traditional ecommerce platforms.

If product discovery increasingly begins on social media, marketplaces such as Amazon and Myntra risk losing their position as the starting point of the shopping journey.

Consumers may still complete purchases on these platforms, but purchase intent is increasingly being shaped elsewhere — through creators, communities and short-form video content.

Marketplaces are already adapting.

Amazon has expanded creator-led initiatives through features such as Amazon Live, influencer storefronts and video-based product discovery experiences.

Myntra has invested in content-led shopping through offerings such as Myntra Studio, Myntra Minis and live-commerce formats that blend creator recommendations with commerce.

The convergence highlights a broader industry trend.

As social platforms add shopping tools and affiliate programmes, ecommerce companies are investing in content and creator ecosystems.

The competition is no longer limited to who completes the transaction. It increasingly extends to who influences the purchase decision in the first place.

Are social platforms becoming the new shopping malls?

For decades, shopping malls served as centres of product discovery.

Consumers often arrived with a broad intention to browse, encountered products they were not actively looking for, and made purchases influenced by what they saw along the way.

Discovery and transaction happened within the same environment.

Social platforms are increasingly replicating that dynamic in the digital world.

Instead of store displays and window shopping, consumers are exposed to products through creators, recommendations and short-form videos embedded within their feeds.

The difference is that discovery is now guided by algorithms that personalise what each user sees.

Research points to a broader shift in consumer behaviour.

A Bain & Company report on how India shops online found that peers, communities and content are becoming increasingly important in shaping purchase decisions, particularly among the next wave of online shoppers.

The report also noted strong growth in video consumption across Tier-2 and smaller cities, alongside rising engagement with livestreaming, influencer-led content and visual search tools.

This evolution has implications beyond social media.

If purchase intent is increasingly formed through content and creator recommendations, traditional ecommerce platforms risk losing their role as the primary starting point of the shopping journey.

Consumers may still complete transactions on marketplaces, but the discovery process is increasingly taking place elsewhere.

The comparison with shopping malls is not perfect, but it captures an important shift.

Social platforms are no longer simply places where people consume content. They are increasingly becoming places where consumers discover products, evaluate options and develop purchase intent — functions that once belonged primarily to retailers and marketplaces.

Where this leaves us

The shift underway is not simply about a new advertising format or a growing preference for short-form video.

It reflects a broader change in how consumers discover products online.

Rather than beginning with a search query or a visit to a marketplace, product discovery is increasingly taking place within content feeds, where creators, communities and algorithms shape consumer interest long before a purchase decision is made.

This has implications across the digital economy.

Brands are adapting their marketing strategies, creators are becoming a more integral part of the commerce ecosystem, and platforms are investing in tools that bring discovery and transactions closer together.

In the process, the boundaries between content, advertising and commerce are becoming increasingly blurred.

Whether social platforms eventually become full-fledged shopping destinations remains an open question.

What is becoming harder to dispute, however, is their growing influence over how purchase intent is formed.

If shopping malls once served as centres of discovery in the physical world, social platforms are increasingly emerging as their digital equivalent.

The battle is no longer limited to who completes the transaction. Increasingly, it is about who shapes purchase intent in the first place.

That may prove to be the most valuable position in the modern digital economy.

Source link

Self-improving AI systems are emerging, but not in the way people expected

by Hansraj Agrawal | Jun 5, 2026 | Technology

Artificial intelligence that can improve itself has long been viewed as a distant possibility. Today, it is beginning to emerge in fragments across the technology industry, not as a single breakthrough but as a series of overlapping shifts in how AI systems are built, trained and deployed.

Anthropic’s latest research offers some of the clearest evidence yet of how far this transition has progressed. The company describes the concept as recursive self-improvement: a process in which AI systems contribute to building better versions of themselves.

Fully autonomous self-improving systems remain some distance away. Yet many of the foundations required for such systems are already visible inside real-world AI development pipelines.

AI enters its own development loop

For decades, AI development followed a relatively straightforward structure. Humans defined problems, wrote code, trained models and evaluated outcomes. Every iteration depended on human effort, limiting progress to the pace at which teams could move through the cycle.

Anthropic’s findings suggest that structure is beginning to change.

The company reports that more than 80 per cent of the code it now merges into its codebase is written by its own AI models. This is not a marginal shift. It means AI is already responsible for the majority of the technical work involved in building AI systems.

The implications extend beyond productivity.

When AI-generated code becomes part of the systems used to train, improve or deploy future models, the technology enters a feedback loop. AI is no longer simply being developed. It is helping shape the conditions that influence its own future development.

The role of engineers is changing alongside it. Rather than writing code line by line, developers increasingly define objectives, review outputs and guide system behaviour. The focus shifts from execution to oversight.

Speed, scale and compounding gains

The most immediate impact of this transition is acceleration.

Anthropic says output per engineer has increased roughly eightfold since 2024. Internal estimates suggest productivity gains of around four times across teams. These figures are significant not simply because they reflect greater efficiency, but because they create compounding effects.

As AI takes on more of the development process, it speeds up the work required to build better AI systems.

Faster code generation enables faster experimentation. Faster experimentation leads to quicker model improvements. Better models, in turn, accelerate development even further.

The result is a feedback loop where each iteration helps improve the next.

External research from Model Evaluation and Threat Research (METR) points to a similar trend. According to its findings, the complexity of tasks that frontier AI models can handle has been doubling roughly every seven months.

Anthropic’s own data suggests these systems are also becoming capable of operating independently for longer periods. Tasks that once took only minutes now extend to hours and, increasingly, full work sessions.

From coding assistant to experimental researcher

The next stage of this evolution goes beyond code generation.

Anthropic describes emerging systems capable of proposing hypotheses, conducting experiments, evaluating outcomes and iterating towards better solutions.

This is important because it mirrors the basic mechanism required for self-improvement. Any system capable of improving itself must be able to generate changes, test them and determine whether those changes actually represent progress.

In one example cited by Anthropic, AI agents working on a research problem recovered nearly the entire performance gap between a baseline and an optimal solution, outperforming human researchers operating under tighter time constraints.

The system was not simply following instructions. It was navigating a defined problem space with a degree of independence.

Yet the limits remain equally important.

Humans still define the objectives, establish evaluation criteria and determine which problems are worth solving. AI operates within a framework created by people.

In other words, current systems are demonstrating the mechanics of self-improvement, but only within boundaries established by humans.

They can optimise within a problem. They cannot yet determine what the problem itself should be.

Microsoft’s approach: Controlled self-improvement

While Anthropic focuses on how self-improving behaviour is emerging naturally inside AI development, Microsoft is attempting to formalise the process.

Its Frontier Tuning initiative aims to create systems that continuously improve through real-world usage while remaining within tightly controlled environments.

At the centre of the approach are reinforcement learning environments where AI models learn from an organisation’s own workflows, data and evaluation signals.

These environments function as controlled spaces where models can improve over time without directly affecting production systems.

The principle is straightforward: learning should not stop once a model is deployed. Instead, systems continue adapting based on how they are used, becoming increasingly specialised to specific organisational needs.

What differentiates Microsoft’s approach is its emphasis on governance.

The learning loop exists, but it operates within compliance frameworks, access controls and predefined evaluation systems. This makes the process predictable, measurable and auditable.

Microsoft is not building fully autonomous self-improving AI. It is building constrained self-improving systems where the feedback loop exists but remains carefully managed.

Google focuses on the missing pieces

While Anthropic and Microsoft are demonstrating what is already possible, Google DeepMind is concentrating on the problems that still need to be solved.

In an interview with Axios, DeepMind chief executive Demis Hassabis identified two major barriers to fully autonomous self-improving systems

The first is the absence of robust world models. For a system to improve itself effectively, it must understand how its actions translate into outcomes. In environments such as chess, this is relatively simple because the rules are fixed and consequences are clear. Real-world environments are far more complex. Outcomes are uncertain, conditions change constantly and cause-and-effect relationships are often difficult to model.

The second challenge is verification. Even if a system generates a potentially better solution, it still needs a reliable way to determine whether that solution actually represents an improvement.

In coding, tests can provide answers. In mathematics, proofs can serve as verification. In many real-world situations, however, there is no immediate or objective signal that defines success. These limitations help explain why self-improving behaviour is emerging unevenly.

Progress is advancing rapidly in areas such as software development and research workflows, where outcomes can be measured precisely. It remains far more limited in environments where success is harder to define.

The rise of organisational learning loops

Even within these constraints, self-improving systems are already beginning to affect businesses.

As AI becomes embedded in workflows, it generates outputs that can be evaluated, refined and fed back into future iterations.

This creates feedback loops not only for AI models but also for organisations themselves.

Over time, companies can build systems that continuously improve based on their own operational data and internal processes.

The competitive implications are significant.

Historically, companies gained advantages through better tools, larger workforces or greater financial resources. Increasingly, advantage may come from how effectively organisations create and manage their own AI learning loops.

Two businesses using the same underlying model can achieve very different outcomes. One may treat AI as a static tool. Another may continuously refine it using internal data, feedback mechanisms and evaluation systems. Over time, the gap between those approaches could widen substantially.

This also changes the nature of software.

Rather than existing as fixed systems updated periodically, AI-driven systems become dynamic and adaptive. Their performance depends not only on how they were designed but also on how effectively they learn from usage.

What comes next?

The developments described by Anthropic, Microsoft and Google point towards a common direction. Self-improving AI is not arriving as a single, dramatic event. Instead, it is emerging gradually through a series of feedback loops embedded inside development pipelines, enterprise systems and research environments.

For now, those loops remain constrained.

Humans still define objectives, establish boundaries and determine what success looks like. AI systems can optimise within those frameworks, but they cannot yet create entirely new ones.

That distinction remains crucial.

Yet the underlying trend is becoming harder to ignore.

AI is increasingly participating in the processes used to build, refine and improve future AI systems. The feedback loops are already forming. The question is no longer whether self-improving AI is possible. It is how far these early forms of self-improvement can evolve before the remaining technical barriers begin to fall.

Source link

Computers were built for humans. Big Tech is redesigning them for AI

by Hansraj Agrawal | Jun 5, 2026 | Technology

For decades, every major breakthrough in computing has been built around a simple premise: A human sits at the centre of the machine.

Whether using a mainframe, a personal computer, a smartphone, or the internet, people have always been the active participants. Humans initiated actions, searched for information, opened applications, and decided what happened next. Computers responded.

That assumption is beginning to change with the rise of artificial intelligence (AI) agents.

A new generation of AI systems can increasingly reason, plan, execute tasks, monitor information, and interact with software systems on behalf of users. Rather than waiting for instructions, these systems can operate continuously in the background, pursuing objectives defined by humans.

As a result, technology companies are redesigning everything from computer chips and operating systems to cloud platforms and search engines for a future in which software increasingly works on behalf of humans rather than waiting for commands.

This is why Nvidia chief executive Jensen Huang believes AI agents will become the “largest consumers of computing resources” and why Qualcomm chief executive Cristiano Amon has described 2026 as “the year of agents”.

The shift, if realised, could become as significant as the move from desktop computing to mobile computing.

How have computers always been built around humans?

Every major computing era has been defined by the way humans interacted with machines.

In the mainframe era, users entered commands into terminals and computers processed them. The personal computer era replaced command lines with graphical interfaces, allowing users to click icons, open files, and navigate software visually.

The internet expanded access to information, but retained the same basic model. Humans searched for information, clicked links, and navigated websites manually.

The smartphone era compressed these interactions into applications. Users tapped icons, opened apps, entered information, and moved between services to complete tasks.

Across each of these transitions, one principle remained unchanged: humans were the active participants and computers were reactive systems.

AI agents reverse that relationship.

Instead of waiting for instructions at every step, software can increasingly operate independently, carrying out tasks and making decisions within boundaries defined by the user.

Why do AI agents change the rules of computing?

The significance of AI agents lies in what differentiates them from traditional software.

They are not simply chatbots that respond to prompts, nor are they digital assistants limited to answering questions. Agentic systems can reason, plan, execute tasks, monitor information sources, coordinate software services, and perform multi-step workflows with limited human involvement.

Consider travel planning.

Today, a user might search for flights, compare prices, monitor fare changes, coordinate schedules, and make bookings manually across multiple websites and applications.

An AI agent could potentially handle those steps autonomously, monitoring fares over time, evaluating options against user preferences, coordinating with calendar availability, and completing bookings once specified conditions are met.

The implications extend far beyond travel. Similar systems are being developed for research, software development, financial analysis, customer support, enterprise workflows, and more.

If software begins acting on behalf of users rather than merely responding to requests, the computing infrastructure supporting it must change as well.

If machines become users, why must chips change?

The first signs of that transformation are appearing in semiconductors.

The initial wave of generative AI was powered primarily by graphics processing units (GPUs), which proved highly effective for training large language models and running inference workloads.

Agentic systems, however, introduce new requirements.

Unlike chatbots, agents operate continuously. They interpret goals, gather information, execute code, coordinate application programming interfaces (APIs), and maintain context across multiple tasks. These workloads require significantly higher concurrency, memory bandwidth, and persistent processing capabilities.

Hardware designed around intermittent human interactions is not optimised for such workloads.

This challenge has prompted chipmakers to rethink processor architecture.

Nvidia recently introduced Vera, which the company describes as the first high-performance datacentre CPU built specifically for agentic AI workloads. Nvidia claims the processor delivers up to 1.8 times the performance of leading x86 datacentre CPUs that have historically powered enterprise computing.

Despite being a new entrant to the CPU market, Nvidia has already secured support from companies including OpenAI, Anthropic, xAI, Oracle Cloud Infrastructure, and ByteDance.

This transformation is also reaching consumer devices.

Apple’s M-series silicon architecture combines CPU, GPU, Neural Processing Unit (NPU), and unified memory into a single design optimised for AI processing. The latest M5 family places particular emphasis on memory bandwidth and on-device AI execution.

Within the Windows ecosystem, Qualcomm’s Snapdragon X platform and Nvidia’s RTX Spark architecture are pursuing similar goals. Nvidia’s platform combines a Blackwell GPU, a 20-core Arm-based CPU, and up to 128 GB of unified memory, delivering memory bandwidth of up to 300 GB/s.

For Qualcomm, the transition extends beyond PCs. The company envisions smartphones, laptops, vehicles, and smart glasses functioning increasingly as endpoints connected to persistent AI systems operating across local and cloud environments.

Why may apps matter less in an agent-driven world?

The rise of AI agents also challenges one of computing’s most enduring concepts: The application.

For decades, software has been organised around applications because humans needed visual interfaces to interact with digital systems. Operating systems existed largely to help users launch, switch between, and manage these applications.

AI agents do not necessarily require those interfaces. Instead, they can interact directly with software services through APIs, retrieving information, executing transactions, and coordinating workflows without opening traditional applications.

As a result, technology companies are exploring agent-first operating systems where applications become backend execution layers while agents serve as the primary interface.

Google’s vision for Android 17 shows how this transition could reach billions of smartphone users. The upcoming version of Android is expected to deepen Gemini’s integration into the operating system, allowing AI systems to understand on-screen context, access system-level functions, and coordinate actions across multiple apps.

Rather than requiring users to manually switch between applications, Google is increasingly positioning Gemini as an orchestration layer capable of completing tasks that span messaging, calendars, maps, shopping, travel, and productivity services.

The broader goal is to make the smartphone less app-centric and more intent-centric. Instead of opening several apps to organise a trip, schedule a meeting, or compare products, users could simply describe an objective and allow an AI system to determine which services need to be accessed and in what sequence.

In this model, applications remain important, but increasingly operate behind the scenes as execution engines rather than primary user interfaces.

Microsoft’s Project Solara offers a glimpse of a similar future.

Described as a chip-to-cloud platform designed for a multi-agent world, Solara is built around a lightweight operating environment that coordinates local and cloud-based agents rather than traditional desktop software.

To demonstrate the concept, Microsoft introduced two reference designs: A wearable smart badge aimed at frontline workers and a desk companion device designed to provide persistent access to enterprise AI systems.

Underlying both is the concept of “Just-in-Time” user interfaces, where generative AI dynamically assembles interfaces according to context rather than relying on pre-built application layouts.

In effect, the interface itself becomes generated on demand.

How are search, cloud and the internet being rebuilt?

The transformation extends beyond devices and operating systems.

For decades, the internet has operated on a reactive model. Users searched for information, opened websites, gathered data, and repeated the process whenever new information was needed.

Agentic systems introduce a different model.

Rather than asking questions repeatedly, users define objectives and allow software systems to pursue them continuously. An AI agent could monitor competitors, track regulatory filings, compare products, follow market developments, or gather research while the user is offline.

Google’s overhaul of Search reflects this shift.

The company is repositioning Search from a tool that helps users locate information towards one that increasingly interprets, summarises, monitors, and delivers information directly.

Gemini, Google’s foundational AI model, is simultaneously becoming an intelligence layer across Search, Android, Gmail, Chrome, Workspace, and other products.

The goal is to move computing from episodic interactions towards continuous execution.

What changes if computers start working for us?

The implications extend beyond technology infrastructure.

If AI systems become capable of coordinating software autonomously, the importance of individual applications could diminish. Users may increasingly interact with a single AI layer rather than managing dozens of separate apps.

Search behaviour could also change fundamentally. Instead of directing users to websites, AI systems may increasingly gather, synthesise, and present information directly.

That shift raises questions for publishers, creators, and businesses that rely on search traffic.

Work itself could be reshaped. Research, scheduling, reporting, coding, customer service, and administrative tasks may increasingly be delegated to software operating in the background.

Even devices could evolve. Smartphones and PCs may become gateways into persistent AI systems rather than destinations where work is performed.

These are the questions increasingly occupying technology executives because they determine not only how computing evolves, but also where value is created across the digital economy.

Why may the transition take longer than Big Tech expects?

Despite the momentum, significant obstacles remain.

Cost remains one of the largest challenges. Agentic workloads require substantial computing resources, making large-scale deployment expensive.

Reliability presents another hurdle. AI systems continue to generate inaccurate outputs and hallucinations, limiting their ability to operate autonomously in sensitive environments.

Security concerns are equally significant. Systems capable of taking actions on behalf of users introduce new risks if compromised or manipulated.

Regulatory questions also remain unresolved. As AI systems gain greater autonomy, questions around accountability, liability, and oversight become increasingly complex.

These constraints suggest the transition may unfold more gradually than some industry leaders anticipate.

The next computing shift

For decades, computers have been tools that waited for human instructions.

The technology industry’s largest companies are now betting that the next era of computing will look different: Machines acting on behalf of people, interacting with other machines, and operating continuously in the background.

How quickly that vision becomes reality remains to be seen, but one thing is clear: Technology companies are redesigning the foundations of computing for a future in which AI systems may become a major new consumer of computing resources alongside human users.

Source link

AI no longer speculative technology but an operational reality: CJI

by Hansraj Agrawal | Jun 5, 2026 | Technology

Artificial intelligence is no longer a speculative technology but an operational reality and poses one of the most significant tests for international law, Chief Justice of India Surya Kant has said, underlining that choices made during this decade will shape the future relationship between technology, power, freedom, and justice.

He also stressed that technology itself is neither inherently benevolent nor inherently harmful.

Speaking at a public lecture in Birkbeck College of University of London on “Artificial Intelligence and International Law”, he said unlike previous technological revolutions, AI does not merely enhance human capacity; it increasingly participates in decision-making processes that were historically considered uniquely human.

“Technology itself is neither inherently benevolent nor inherently harmful. Its impact depends upon the legal, political, and ethical frameworks within which societies choose to deploy it. The responsibility of law, therefore, is neither to resist technological progress nor to surrender unquestioningly before it. Its responsibility is to ensure that technological power remains accountable to constitutional values, democratic legitimacy, and human dignity,” he said.

CJI Kant said AI is now an operational reality that is reshaping governance, commerce, warfare, communication, public administration, and increasingly, the exercise of judicial and sovereign power itself.

“Governments now utilise algorithmic systems to allocate welfare benefits, assess immigration applications, monitor borders, regulate financial systems, and support policing functions. Militaries are rapidly developing autonomous capabilities. Courts across jurisdictions are beginning to confront questions involving AI-generated evidence, automated decision-making, and digital due process. Private corporations possess technological capacities that rival, and in some instances exceed, the informational reach of sovereign states,” he said.

Artificial intelligence poses one of the most significant tests for international law in its modern evolution, he said, adding choices made during this decade will shape the relationship between technology, power, freedom, and justice for generations to come.

“The central challenge before us is to ensure that, in an age of intelligent machines, humanity retains authorship of the principles by which it is governed. If international law can rise to that challenge, artificial intelligence may become not merely a technological revolution, but an opportunity to reaffirm the values that lie at the foundation of democratic civilisation itself,” he underscored.

CJI Kant, who is on a six-day UK trip, said, artificial intelligence presents unprecedented opportunities for strengthening the administration of justice and across jurisdictions, courts are increasingly leveraging AI-driven tools to assist with legal research, case management, translation services, transcription of proceedings, document classification, and the identification of judicial precedents.

“When deployed responsibly and under appropriate human supervision, such technologies can help reduce delays, improve efficiency, expand access to legal information, and enable judges and court administrators to focus their attention on the more nuanced and inherently human aspects of adjudication. AI, therefore, should not be viewed solely as a source of legal complexity, but also as a powerful instrument for advancing the constitutional promise of timely, accessible, and effective justice,” he said.

The CJI wondered whether AI will influence international law as the transformation is already underway and the real question is whether the existing architecture of international law possesses the conceptual elasticity necessary to absorb this disruption.

“We need to assess if the fundamental doctrines of international law, namely sovereignty, human rights and enforceability of foreign awards/decrees will be able to adapt sufficiently to govern algorithmic power? Or are we approaching a moment that requires an entirely new legal imagination?” he said.

The traditional international law is deeply anchored in territoriality and AI fundamentally unsettles these assumptions, the CJI said, adding AI systems function through globally distributed architectures that frequently transcend territorial boundaries altogether.

“A model may be trained on datasets collected across multiple jurisdictions, refined through computational infrastructure situated elsewhere, deployed through cloud-based systems spanning several continents, and ultimately produce decisions affecting individuals far removed from every point in that chain,” he said.

Thanking Birkbeck college for hosting this important conversation, CJI said at moments of profound technological transformation, dialogue between courts, universities, governments, and civil societies becomes indispensable.

“Ultimately, the future of artificial intelligence will be shaped not only by innovation but by the legal and moral choices that humanity collectively chooses to make,” he said and highlighted that the challenge before the international community is not merely to regulate technological capability, but to preserve legal responsibility in environments where decision-making is increasingly mediated through algorithmic systems.

He added that if responsibility becomes too fragmented to identify, accountability itself risks becoming illusory.

Source link

HMD Vibe 2 5G review: Democratises AI, but the smartphone shines elsewhere

by Hansraj Agrawal | Jun 5, 2026 | Technology

Most budget phones today follow a very predictable formula. You get a big battery, a decent display, a usable camera, and enough performance to get by. Beyond that, there’s usually nothing that really stands out.

The HMD Vibe 2 5G mostly fits into that same category on paper. It ticks the usual boxes with a display of 120Hz refresh rate, large capacity battery, clean software, and 5G support. But what makes it interesting is not the hardware. It’s the fact that this is the first smartphone in the country to come pre-loaded with the Indian AI startup Sarvam AI’s Indus assistant.

That immediately changes how you look at the device. Because now, the question is not just how good the phone is, but also whether this AI-first approach actually adds anything meaningful to the experience.

I’ve been using the Vibe 2 5G as a daily device for a while now, and the answer ends up being a mix of both expected and unexpected.

Design and build

The Vibe 2 5G doesn’t feel like a typical sub-Rs 12,000 phone the moment you pick it up. The design leans closer to what you would expect from mid-range devices, with a clean finish and a minimalistic look. The back panel has a simple design without trying to do too much, and the camera module sits fairly flush, which avoids the usual wobble when placed on a flat surface.

It is on the heavier side at around 210g, but that largely comes from the 6,000mAh battery inside. In hand, it doesn’t feel cheap, despite its all-plastic construction.

You also get an IP64 rating, which is not very common at this price point, and a side-mounted fingerprint sensor that works reliably. There’s even a dedicated microSD slot along with dual SIM support, which is something a lot of users still care about.

Overall, the design doesn’t try to stand out—it just avoids obvious compromises, and that works in its favour.

The experience

The display is a 6.7-inch IPS LCD panel with a 120Hz refresh rate. It is only HD+, which might look like a compromise on paper, but in actual use, it doesn’t feel as limiting for everyday tasks like scrolling, social media, or video consumption. The higher refresh rate helps more than resolution here, making the phone feel smooth. The display is also bright enough for outdoor use; however, viewing angles aren’t great, as the display appears darker when viewed from the sides.

Performance is handled by the Unisoc T8200 chip, and for a budget device, it holds up better than expected. Day-to-day usage—apps like WhatsApp, YouTube, browsing, and even light gaming—runs without major issues. However, it can get laggy when you have multiple apps running in the background, especially given the 4GB RAM.

Thermals are also under control. Even during longer sessions, the phone doesn’t heat up noticeably, which is something budget devices often struggle with.

Battery life is one of the stronger aspects here. The 6,000mAh battery easily lasts through a full day and often stretches into the next, depending on usage. Charging is limited to 18W, which is not particularly fast, but acceptable at this price—especially since the charger is included in the box.

Audio is where compromises start to show. You only get a single speaker, which is fine for calls and casual use, but not ideal for media consumption. The presence of a 3.5mm headphone jack does help balance that out.

On the software side, the phone runs Android 16 with a clean UI and minimal bloatware. That alone makes a noticeable difference in daily use. It feels lighter and more responsive compared to heavily skinned alternatives in this segment.

The camera setup is basic, which is expected at this price. You get a 50MP primary sensor along with a depth sensor. In good lighting, the camera manages to capture decent shots with acceptable detail and slightly boosted colours. It’s not overly processed, but you do notice the limitations when you start looking closely.

Low-light performance is average. Images tend to lose detail, and noise becomes visible fairly quickly. There is a night mode, but it only helps to an extent.

The front gets an 8MP sensor, and while it is good enough for video calls and occasional daylight selfies, images tend to lose sharpness and soften facial details.

Video recording goes up to 4K at 30fps, which is a good addition on paper for this segment. The output is usable for casual clips, but stabilisation isn’t the strongest, so you do need steady hands.

Indus AI and other intelligent features

The Indus AI assistant is easily the most interesting part of this phone, but also the most inconsistent.

The interface is familiar. It is comparable to similar apps like ChatGPT or Google Gemini, so there’s no real learning curve if you are familiar with any of the new age AI assistant apps. You can type prompts, ask questions, and get responses in a conversational format.

For basic use, it works reasonably well. General questions, simple explanations, and everyday queries are handled without much trouble. Where it starts to fall short is when you go deeper.

In many cases, the responses feel more like summaries of web results rather than properly reasoned answers. It can retrieve information, but doesn’t always build on it effectively. For example, while exploring a physics concept like “Boltzmann Brain”, when I gave Indus a weak or incorrect premise, it often did not challenge or correct it. That said, this wasn’t always the case in more common scenarios. For instance, when I asked why laptops don’t have more than 8GB RAM, it did correct the premise. So the behaviour isn’t entirely consistent.

To be fair, this is still a beta version of the AI assistant and is based on Sarvam’s 105B parameter model, which is significantly smaller than the models used by leading global AI platforms. So some gaps are expected at this stage.

Where Indus actually stands out is language support. It handles Hinglish and mixed-language inputs surprisingly well. You can switch between Hindi and English, use local terms, and it still understands context correctly. That level of localisation feels more natural compared to most global assistants.

Voice input also works, but with a slight delay. The app first transcribes the audio and then processes it as text. It’s accurate, but not as seamless as real-time voice assistants.

The bigger issue is integration. Indus exists as a standalone app, and while you can set other AI apps like ChatGPT as your default assistant, Indus does not show up in the list of available assistants at all. This is likely due to the lack of native voice support.

So while the idea is interesting, it doesn’t yet feel like something that changes how you use the phone.

Beyond the Indus app, the phone gets a system-wide search bar in the app drawer that the company calls AI Search. It works as a quick lookup tool for apps, stored files, and even documents, all from one place. In regular use, it feels faster than manually browsing through folders or scrolling through the app list. It also suggests apps based on what you might be looking for, which adds a bit of convenience.

Verdict

The HMD Vibe 2 5G is a good budget phone, and that’s mostly because it gets the basics right. The performance is stable, the battery lasts long, the display feels smooth, and the software is clean.

The Indus AI assistant adds a different angle to the device, but right now, it feels more like an early experiment than a fully integrated feature. It works, but it doesn’t yet become a core part of the experience.

If you’re looking for a reliable budget phone under Rs 15,000, this makes a strong case. The AI angle is interesting, especially for Indian users, but it’s not the reason to buy the phone—at least not yet.

Price: Starts at Rs 12,999

Source link

« Older Entries

Next Entries »

Times Top News

Times Top News