AI’s Social Layer & Sarvam’s Token Factory

AI’s Social Layer & Sarvam’s Token Factory
AI social layer

In the web era, we operated and connected through distinct platforms, each catering to a specific user need. Search engines were for information, social networks for people, marketplaces for shopping and messaging apps for conversations.

But AI assistants have begun to disrupt it all, including the web itself. Today, anyone looking for a new phone may no longer need to search Google, review websites or compare products across multiple tabs. Instead, they can turn to an AI assistant to evaluate options, compare prices and get personalised recommendations. 

The next stop for AI assistants could very well be social media platforms that have long controlled online discovery through their proprietary algorithms. 

As AI increasingly takes over product discovery, recommendations and decision-making, platforms that depend on search traffic, user attention and algorithmic distribution could find their influence under threat. But what does it really mean for startups reliant on clicks, engagement and advertising? And is AI entering a social era? Let’s answer these questions in today’s edition of The AI Shift.

The Social Feed Is Losing Its Grip

Social media giants such as Facebook, Instagram, X and TikTok built their dominance around algorithm-driven feeds, determining what users discover, engage with and buy. 

Whether it is a friend’s post, a creator’s video, a trending topic or a product recommendation, these platforms have largely controlled what appears on users’ screens. AI assistants, however, are beginning to challenge this model by putting user intent ahead of algorithmic discovery. 

Instead of scrolling through multiple feeds or websites, users can simply ask an AI assistant what they want to know or buy. They can then get recommendations based on reviews, pricing, specifications and personal preferences.

Kushal Bhagia, founding partner at All In Capital, believes traditional discovery mechanisms and top-of-funnel platforms could be among the biggest casualties if AI assistants become the primary interaction layer.

His argument is that much of the social internet depends on curated feeds, algorithmic discovery and SEO-driven aggregation. If an AI assistant can retrieve, summarise and contextualise information directly, users may avoid visiting websites.

“Businesses whose primary value is helping users discover something could face pressure,” Bhagai said.

The shift is already visible in ecommerce. With the rollout of Businesses in ChatGPT, OpenAI now allows eligible businesses to claim their profiles and share product information. Eligible ecommerce and online-service businesses in the US can also create Business Agents that interact directly with customers. 

If adoption in India continues to grow beyond its current level, then OpenAI is very likely to offer this to India and other geographies. With businesses talking directly to users through these agents, the social layer may very well lose its place in today’s commerce value chain. 

AI social layer

AI In The Social Circle

Having said that, AI assistants may not replace social media platforms altogether, but they are fundamentally changing how people interact online. 

Social media platforms have traditionally connected users with friends, communities and content creators, and AI chatbots are increasingly becoming spaces for conversations, companionship and even emotional support.

WARC’s 2026 Consumer Trends research points to the growing use of AI chatbots for social fulfilment, including companionship and mental-health support.

This emerging trend is also backed by academic research. A study published in Nature Human Behaviour in August 2026 examined 1,131 US adults who used Character.AI, alongside thousands of chat sessions. It found that users with smaller offline social networks were more likely to report companionship as their primary reason for using AI chatbots.

Another study published in the same journal in September analysed more than 54,000 posts from Replika and ChatGPT communities. It found evidence of emotional attachment to AI systems, especially when the chatbots changed in ways they disliked.

These findings suggest that AI assistants are beginning to occupy a space that has traditionally belonged to those close to users, a job that was thus far held by social networks, messaging platforms and personal relationships.

This means that social media could increasingly serve as a destination for entertainment, content consumption and community engagement, while AI assistants become a more personalised interface for conversations, advice and decision-making.

In effect, AI assistants could evolve from tools that help users find information into platforms that people turn to for interaction, companionship and guidance.

When AI Agents Take The Wheel

The bigger shift could come when AI assistants go beyond recommendations and start taking action on behalf of users. For instance, an AI agent could book flights, compare insurance policies, negotiate purchases or analyse investment options. Instead of visiting multiple websites and apps, users could simply tell an AI agent what they want and let it handle the process.

Today, most online activities require people to search, compare, click, fill out forms and complete transactions themselves. AI agents could take over many of these tasks, changing how businesses interact with their customers.

OpenAI’s Dots offers a glimpse of this future. These always-on AI agents can connect to thousands of applications, work towards users’ goals and learn their preferences over time. One can imagine a similar future for other personal agents like Muse or Instinct. 

And for AI startups, this shift presents both challenges and opportunities.

Businesses that rely heavily on social media algorithms, advertising and user engagement to attract customers could face pressure. However, startups with loyal communities, unique content and strong customer relationships may be better placed to retain users.

Anuj Gupta, founder and CEO of KiteFishAI, an AI startup that builds compact open-weight small language models, sees opportunities emerging in areas such as agent-to-agent commerce, AI-driven communities, digital identities and systems that establish trust between autonomous AI agents.

Meanwhile, Bhagia expects businesses to start earning revenue through personalised recommendations, transaction fees and commerce conducted by AI agents. Simply put: companies may soon have to compete not just for the attention of people, but also for the attention of AI agents acting on their behalf.


The Weekly Buzz: Sarvam’s Token Factory Takes Aim At India’s AI Inference Market

Sarvam AI has launched Token Factory, an inference platform that allows developers and enterprises to access multiple AI models through a single API, without having to build and manage their own GPU infrastructure. In simple terms, Sarvam is offering the computing layer that runs AI models and generates responses, with inference taking place on GPUs operated by Sarvam in India. The platform includes Sarvam 105B alongside open models such as DeepSeek, GLM, Kimi and Gemma. 

The bigger proposition is around cost, speed and localisation. Sarvam says its serving stack uses techniques such as cache-aware routing, separate prefill and decode infrastructure and multi-tier caching to make inference faster and more efficient. It is also offering serverless and dedicated capacity, giving enterprises different ways to consume the infrastructure. For Indian businesses handling sensitive workloads, the fact that inference can happen within India could be another draw, particularly as data residency and compliance become bigger considerations.

However, Sarvam is entering a market that already has both Indian and global competition. OpenRouter, recently acquired by Stripe, has built a popular model-routing layer that gives developers access to hundreds of models through a unified API, while Indian players such as JarvisLabs and Simplismart are building their own inference offerings. JarvisLabs focuses on simplifying access to open models, whereas Simplismart has pushed further into model serving, optimisation and dedicated infrastructure. Sarvam’s differentiation therefore rests partly on combining its own models, India-based infrastructure and an inference stack optimised for its workloads.

The bigger challenge will be proving that this advantage translates into better economics and reliability at production scale. Model access itself is becoming increasingly commoditised, and developers can switch between providers relatively easily. Sarvam will need to show that its infrastructure can consistently deliver competitive latency, token costs and model availability while handling large enterprise workloads. If it can, Token Factory could become more than another model API and potentially form part of the infrastructure layer powering India’s emerging AI application and agent ecosystem.


Top Stories From India & Around The World

  • OpenAI Brings Inference To India: The program will allow businesses to process supported GPT-6 Luna workloads within the country. The move addresses data localisation requirements that have been a barrier to AI adoption among regulated businesses and increases competition for Indian AI companies offering locally hosted models.
  • Bharat Semi Raises $24 Mn: The Bengaluru-based semiconductor startup’s seed round was led by Info Edge Group, South Park Commons and Indus Bridge Ventures. It will use the capital to develop low-power, low-latency sensors for physical AI applications and expand its engineering teams.
  • MeitY Reviews Physical AI Data Collection: The ministry has sought details from Pronto, Snabbit and Urban Company over their collection and use of gig workers’ video and sensor data for training physical AI models, highlighting concerns around worker consent, consumer privacy and the transfer of Indian training data overseas.
  • AI Giants Face Lawsuit: OpenAI, Anthropic, Google, Meta and SpaceX’s AI division are facing a lawsuit from nonprofit groups seeking to halt the development of increasingly advanced AI systems, arguing that the companies have failed to demonstrate adequate control over frontier models.
  • How Deep Is Nvidia’s AI Ecosystem In India? Nvidia’s AI ecosystem in India is expanding beyond GPU sales, with the chipmaker increasingly embedded across the country’s AI infrastructure, cloud, data centre and startup landscape. 

Startup In The Spotlight: ByteAsk

Founded in 2026 by IIT Delhi alumni Anirudha Kulkarni and Pratyush Saini,, ByteAsk is building an AI coding agent focused on C and C++, two programming languages widely used in mission-critical systems across defence, automobiles, aerospace, medical technology and financial services sectors.

While AI coding tools have largely focused on mainstream software development, organisations working with C and C++ often face additional challenges around performance, reliability, security and legacy code. ByteAsk is designed to help developers in these environments use AI for software development without compromising control over sensitive code and intellectual property.

The platform supports both open- and closed-source AI models and can be deployed entirely on-premise, allowing enterprises to keep their code and data within their own infrastructure. The startup is developing two offerings, ByteAsk Harness and a B2B enterprise product, with a focus on making AI-assisted C/C++ development more reliable rather than simply generating code.

ByteAsk has raised $1 Mn from Y Combinator and follows a subscription-led business model complemented by enterprise services. The startup is targeting the growing AI coding market, estimated at ~11 Bn in annualised value as of April 2026, while carving out a niche around C/C++ and the performance-critical systems where these languages remain central.


Prompt Of The Week

What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? 

Here’s Mashhood Raza Azka, CEO of VoxForce.ai, with a prompt he uses to work out where AI-powered Human Agents can improve customer engagement across digital channels.

“You are an expert in AI-powered Human Agents, customer experience strategy and business process transformation.

Here is our situation: We are a business looking to improve customer engagement across our digital channels. Our customers interact with us for product discovery, sales enquiries, support, appointments and service requests. We want to understand how AI-powered Human Agents can handle these interactions through natural voice and video conversations.

I need you to:

  • Identify customer-facing processes where AI-powered Human Agents can deliver the greatest business value.
  • Recommend suitable AI Agent personas for sales, customer support, reception, product advisory and lead qualification.
  • Map how these agents should understand customer intent, respond to queries and guide conversations towards meaningful outcomes.
  • Identify opportunities to integrate lead capture, appointment scheduling, product recommendations and payment workflows into conversations.
  • Define when an AI Agent should resolve an interaction independently and when it should hand over to a human.
  • Highlight the knowledge, integrations and safeguards required for reliable AI Agent performance.
  • Recommend measurable KPIs to assess engagement, customer satisfaction, lead conversion and operational efficiency.

Prioritise recommendations based on business impact, implementation feasibility and customer experience. Keep the output specific, practical and execution-oriented.”

Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.

[Edited by Shishir Parasher]
[Creatives by Varshita Srivastava]

The post AI’s Social Layer & Sarvam’s Token Factory appeared first on Inc42 Media.