Human-AI Hybrid: The New Reality For Indian Startups

Introducing The AI Shift by Inc42, our all-new weekly newsletter that delves deep into the world of artificial intelligence, LLMs, big tech giants and the major trends sweeping the Indian startup and tech ecosystem. Here’s the first edition; do send us your feedback and suggestions so we can improve as we go along!

AI has moved decisively beyond pilots and experimentation. For startups today, the question is no longer about whether to use AI, but where its autonomy should stop, and human judgment must take over.

Over the past year, Indian startups have embedded AI across product development, customer support, sales, and internal operations. Development cycles have compressed, costs have fallen sharply, and leaner teams have become the new default mode. In some cases, entire workflows are now run end-to-end by machines.

But as AI systems take on greater responsibility, founders are confronting a hard puzzle. Full automation may scale efficiency, but it also scales risk. When humans are removed entirely from the loop, failures become harder to explain, trust erodes faster, and accountability breaks down. This is why keeping humans in the loop is no longer a choice, even for those who boast of being AI natives.

While in theory AI promises frictionless execution, Indian startups are coming to terms with the fact that blind automation introduces new fragilities.

For Ganesh Gopalan, cofounder and CEO of Gnani.ai, the issue runs even deeper. AI systems, he argues, learn patterns, not meaning. “A system can hear words and still get the intent wrong unless someone has shown it through real examples,” he says.

This is where human-in-the-loop becomes indispensable.

Human feedback during training helps refine models, correct drift, and ground AI systems in cultural and linguistic context before they are deployed at scale. Without this layer, models may perform well in controlled tests but struggle in live conversations, particularly in multilingual, high-context markets like India.

Gopalan said that early human involvement here helps prevent costly errors, protects the brand, and strengthens customer trust.

On the other side, humans with AI help amplify the output.

“AI by itself will probably be able to go about 55%. But a human agent in collaboration does 65%,” says Shayak Mazumder, the CEO of Adya.ai. That incremental improvement is not marginal. It is often the difference between an AI system that merely functions and one that users trust.

For startups operating in trust-heavy or regulated sectors, human oversight is fast becoming the control layer that prevents AI from becoming brittle at scale.

One of the clearest effects of AI adoption in Indian startups is the compression of teams and hierarchy.

According to Neeti Sharma, the CEO of TeamLease Digital, AI is pushing startups toward leaner, capability-led organisation design.

While overall tech hiring remains cautious, demand for AI, data, and cloud roles continues to grow 35-45% year-on-year (YoY). At the same time, 30-35% of tasks in entry-level tech roles can now be automated, explaining slower junior hiring and flatter organisational structures.

The workforce shape that emerges is a barbell model. Very small teams at the early stage, followed by selective, high-impact hiring as products mature and global complexity increases. Middle layers thin out and ownership expands.

This shift is already visible in how products are being built at Indian startups.

YourTribe founder Deepak Subramanian told Inc42 that AI has materially changed the company’s economics. “Our development cost in 2025 was about 40% of what we spent in 2024, and we built more features than we did the previous year.”

What earlier took nearly two months of engineering effort can now be completed in roughly two weeks. But that speed comes with new trade-offs. Developers increasingly write, discard, and rebuild code as AI-generated outputs collide with real-world constraints. Execution accelerates, but judgment becomes more critical.

This efficiency is also enabling an entirely new startup archetype.

At SpeakX, this logic has been pushed to its extreme. The edtech startup’s early profitability has been driven by a tightly run, automation-first operating model. SpeakX operates with a 20-member team and no human-led learning delivery, relying entirely on its tech stack to handle instruction, feedback, and scale.

“With zero human touchpoint, we are a 20-member team, building an end-to-end completely tech product,” said Arpit Mittal, the founder of SpeakX. By removing human-led delivery from the equation, the startup has been able to keep costs low, even as “zero human touchpoint” products still depend on human judgment upstream.

According to Ravi Kaklasaria, the CEO of edForce, the startup growth curve has shifted dramatically in just the last two years. Earlier, startups needed five or six people to get started, scaling to dozens as they grew. Today, a single founder, supported by AI tools, can build meaningful products in weeks.

In some cases, even companies approaching significant valuations operate with fewer than a dozen people. AI agents write code, generate content, and run outbound workflows continuously.

But Kaklasaria draws a clear boundary. When it comes to enterprise sales, CXO relationships, and trust-driven negotiations, human presence remains irreplaceable. “That humanness is still very much relevant,” he said.

So, with speed and broader skill set requirements, are we witnessing burnout among startups?

Rishi Bal, the head of BharatGen, says the ecosystem is still in a transition phase rather than one facing widespread burnout. He sees two clear camps emerging.

While pockets of over-aggressive adoption exist, he does not see burnout as a broad trend yet, arguing that most teams are still figuring out how to work with AI rather than being overwhelmed by it.

Across these examples, a consistent pattern is emerging — ‘human-in-the-loop’ as the control layer.

According to Bal, the concept of humans-in-the-loop particularly refers to providing AI in the form of co-pilots that make recommendations, but in the end, humans must make the final decision. For Indian startups navigating regulation, trust, and diverse user bases, this hybrid architecture is increasingly a strategic choice.

And for India’s startup ecosystem, the implications are structural rather than incremental. Startups are being built with fewer people, lower burn, and shorter iteration cycles. But they are also being built with sharper definitions of responsibility. AI is compressing the execution timeline and making it happen, whereas humans are shifting focus to judgment. All in all, while we may see AI running the Indian startup engine in the not-so-distant future, humans will be at the wheel.

Heated debates on AI are nothing new for X or LinkedIn, but the recent exchange between Zoho founder Sridhar Vembu, and Y Combinator president Garry Tan was a duel of heavyweights.

The two clearly seem to have differences on what “vibe coding” can achieve and whether it will disrupt the traditional horizontal SaaS model. To begin with, Tan took a dig at Zoho’s business being one of the first to be hit by vibe coding, as non-technical users start building custom software using platforms like Replit, Taskade, and Emergent Labs. He argued that users wouldn’t pay for a per-seat monthly subscription for Zoho, and instead would build an entire replacement over the weekend.

The former Zoho CEO and current chief scientist responded by citing Zoho’s current customer growth of over 50% and questioned why no “vibe coding” project hasn’t replaced mission-critical productivity tools like email, spreadsheets, accounting, or messaging. Vembu also put a bet publicly that Zoho will outshine and outlive Tan’s vibe coding companies.

AI will kill SaaS is a popular claim these days, but which side of the debate do you fall on? Will these fly-by-night vibe coding projects actually impact the likes of Zoho and other horizontal SaaS players?

Every enterprise wants to use AI to improve its productivity, efficiency and reduce people costs, but one-size-fits-all solutions rarely work out.

Founded in 2023 by Shayak Mazumder, Archana Mazumder and Angad Singh Ahluwalia, Bengaluru-based Adya AI is building a full-stack AI platform where enterprises can create their own LLMs, agents, applications, memory, governance, and deployment workflows in one place.

In a conversation with Inc42, founder and CEO Mazumder disclosed that close to 10,000 developers are actively building on the platform, with governments, defence bodies, banks, fintechs, energy firms, and enterprises in India and the US as clients.

Adya’s offerings span across four layers: a core IP and research layer solving agent scalability, governance, and memory; a platform layer for building LLMs, agents, and applications; an ecosystem layer enabling developers and partners; and a solutions layer offering composable, enterprise-ready AI systems deployable on any cloud.

In theory, Adya’s platform should simplify AI adoption based on the enterprise needs and allows businesses complete control over their data, infrastructure, and how AI behaves in production.

What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? Here’s Arjun Nagulapally, CTO of AIONOS, with his suggestion that helps him get strategic clarity & options for decision making: “You are my strategic thought partner and a top 1% Fortune‑500 advisor. Here is the situation: [paste context: market, company stage, constraints, KPIs].

Respond in a concise executive brief (headings + bullets) that I can paste into a CXO deck.”

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