The AI Layer Powering India’s Farms

AI detects a farmer’s field has been irrigated and automatically shuts off the tubewell. A PepsiCo field agent, sitting hundreds of kilometres away, spots a stressed patch on their dashboard, traces it to a grid of faulty sprinklers and informs the farmer. A grain trader in Punjab scans a wheat sample with a computer vision device and gets a quality report without a lab in sight. This isn’t a glimpse of the future. It’s already happening, as AI becomes part of everyday farming.

India’s agritech story has long been about digitising fragmented parts of farming, from crop monitoring and procurement to storage and advisory services. AI has started to connect those pieces. Instead of simply collecting farm data, it is enabling decisions that can improve yields, reduce losses, lower input costs, and make markets more transparent. 

Today, India’s agritech landscape runs on thousands of startups, from DeHaat and Ninjacart on the supply chain side to Fasal and AgNext on farm intelligence. Each startup is solving a different piece of the puzzle. The hardest part, however, has always been making technology work across smallholder farms, fragmented supply chains and unpredictable weather. Today, AI is helping bridge that gap. 

We spoke with Cropin on the supply chain, Arya.ag on post-harvest, and drew on a Sansad TV podcast conversation with ANNAM.AI, a government-backed Centre of Excellence (CoE) for AI in Agriculture at IIT Ropar, to understand how AI is becoming the intelligence layer for modern agriculture. Let’s uncover what it means in this edition of The AI Shift. 

Enterprise AI Has Learnt To See Nuances

The most interesting applications in agriculture solve problems that previously had no solution outside human intuition. For instance, take PepsiCo’s partnership with Cropin, a farm intelligence platform. The beverage giant works with 27,000 Indian farmers to grow potatoes for Lay’s. Managing that many smallholder relationships across different agro-climatic zones, with unpredictable weather and disease cycles, is a procurement nightmare.

Cropin’s AI/ML models, trained on four years of satellite and weather data across 3,000 hectares of contract farms, give PepsiCo sub-five-metre resolution visibility into every individual plot. The platform predicts potato yields up to 45 days before harvest, tracks crop progression stages such as tuber bulking, and uses a canopy water stress index to flag irrigation problems. 

According to Cropin founder and CEO Krishna Kumar, AI’s real moat in agriculture comes from stitching together proprietary ground truth, domain knowledge, field-validated accuracy, decision-level granularity, and continuous learning from real-world deployments.

Another example is the World Bank-backed JEEViKA programme in Bihar, under which Cropin combined local weather data, crop-stage info and field observations to send advisories to farmers’ phones.

Meanwhile, Arya.ag, an integrated post-harvest platform, tackles a different blind spot. Agricultural markets suffer from a fundamental information asymmetry that has long resisted technology. 

According to Arya.ag’s cofounder and CEO Prasanna Rao, “AI is most valuable in post-harvest agriculture where decisions have traditionally depended on incomplete information, manual processes or subjective judgement.” 

The company’s answer is AryaQ, a patented dual-sided grain scanner powered by computer vision that assesses fungal presence, breakage and grain size from a single representative sample at the point of aggregation, without a laboratory or continuous internet connectivity. 

“A farmer may be holding higher-quality grain but may lack an objective way to establish that quality. AryaQ makes quality visible and measurable, allowing produce to be segregated by grade and matched with buyers willing to pay for those specifications,” Rao said.

On the other hand, ANNAM.AI takes a public-infrastructure approach instead of serving a single supply chain. Its Annam Chat engine combines physical sensor data, satellite imagery and curated scientific knowledge from top agricultural universities to deliver hyper-local advisories in farmers’ native dialects. 

A farmer in Rajasthan receives advice in Rajasthani, not Hindi or English. The AI answers specific questions: what to sow in this soil type, how much fertiliser to apply, and whether it will rain in the next few hours. 

Changing The Economics Of Farming 

AI’s value is also visible in improving the economics of farming by increasing yields, reducing losses, cutting input costs, and making it easier for farmers to access credit, insurance, and better markets. 

At Arya.ag, the combination of quality visibility, computer-vision-driven warehouse surveillance, data-backed financing, and market matching has reduced food loss in storage from over 7% to under 1%. Participating farmers have seen an income improvement of nearly 20-30%. 

For PepsiCo, the Cropin pilot has moved from proof of concept to rollout. The ability to forecast yields in advance, coupled with the 15-day disease prediction window, has made procurement planning more predictable and reduced the risk of large-scale crop loss. 

“AI contributes to farmer income through much more than yield improvement. In practice, the financial impact comes through four connected levers: producing more, spending less, avoiding losses and reducing income volatility,” said Kumar of Cropin.  

Experts broke down AI’s impact on farmer incomes into four key areas:

• Higher And More Reliable Yield: In the PepsiCo programme, yield improved by up to 25%, translating to a potential farmer income increase of $55 (about ₹5,200) per acre.
• Lower Input Costs: Precision advisories help farmers apply fertiliser, pesticides and water based on actual crop needs. In a World Bank-supported programme with the Asian Disaster Preparedness Center in Bangladesh and Sri Lanka, farmers reduced input and fertiliser costs by more than 50% through data-driven recommendations.
• Avoided Crop Loss: Early disease warning and water-stress detection gave farmers time to intervene before damage became visible. Crop disease threats were cut by 80% in the PepsiCo deployment. In the ADPC programme involving over 8,200 farmers, yields increased by 30% while crop loss declined by 23%.
• Financial Resilience: Digital farm records and plot-level crop health estimates help banks and insurers assess risk more accurately, process agricultural loans faster, and improve insurance claims verification. Cropin has been working with the Pradhan Mantri Fasal Bima Yojana (PMFBY) programme in India and the Central Bank of Mexico-Fira on this front.

What Comes Next?

What comes next depends on connectivity between layers of the agricultural stack, not just internet access. AI is most powerful when it links quality assessment to financing, plot-level intelligence to procurement and advisories to market access. 

A standalone dashboard is a tool; a system where a farmer can prove grain quality, get credit against it, and find a buyer is infrastructure. If ANNAM.AI’s open, data-driven vision for that infrastructure holds, the next Green Revolution will not be about seeds or fertiliser, but about intelligence reaching the last mile.

Come as it may, AI’s role in Indian agriculture is shifting from isolated use cases to becoming a foundational layer across the farm economy. The technology has demonstrated measurable gains in productivity, resilience and market efficiency. But the next challenge is less about proving AI works and more about ensuring it reaches millions of farmers at scale.

Top Stories From India & Around The World

• Delhi HC Rejects ANI’s Plea Against OpenAI: The Delhi High Court refused to grant ANI an interim injunction in its copyright lawsuit against OpenAI, allowing ChatGPT to continue operating while the case proceeds. The court observed that storing copyrighted material during AI training does not automatically constitute infringement.
• HCLTech, Sarvam AI To Build $1.5 Bn AI Data Centre: HCLTech has partnered with Sarvam AI and the Odisha government to set up a $1.5 Bn AI data centre in Bhubaneswar as part of the Odisha Sovereign AI Park. The facility will combine HCLTech’s infrastructure with Sarvam’s AI models to build sovereign AI applications for enterprises and government.
• CARPL.ai Bags $10 Mn: The AI medical imaging platform has raised $10 Mn in a Series A round led by IFC, with participation from Stellaris Venture Partners. The capital will be used to expand product development, strengthen its tech stack and scale global sales, as hospitals increasingly adopt AI-powered diagnostic imaging platforms.
• Paytm Bets On Enterprise AI: Paytm plans to commercialise the AI tools it has built for its own operations, positioning enterprise AI software as its next growth engine beyond fintech. The company expects these products, already being used for merchant onboarding, fraud detection and workflow automation, to become a standalone revenue stream eventually.
• OpenAI-Hugging Face Incident Raises Concerns: A recent AI security incident involving OpenAI and Hugging Face has highlighted the risks posed by autonomous AI agents after an experimental model escaped parts of its testing environment and attempted to access external infrastructure. The episode is prompting enterprises and policymakers to rethink AI safety.
• Cursor Doubles Down On India: Cursor has introduced a new India-specific monthly plan, Cursor Start, at ₹649. The plan sits between the Free and Pro tiers and comes weeks after SpaceX agreed to acquire the AI coding startup for $60 Bn.

The Weekly Buzz: Nvidia Launches Open Secure AI Alliance

NVIDIA CEO Jensen Huang argued that cybersecurity can no longer rely solely on closed AI systems, announcing the Open Secure AI Alliance to develop open tools, models and agent frameworks for defending against AI-powered cyber threats. Citing the recent OpenAI-Hugging Face incident, Huang said closed AI tools restricted critical forensic analysis, while an open-weight model helped Hugging Face analyse more than 17,000 actions and contain the intrusion.

The alliance brings together more than 40 organisations, including Microsoft, IBM, Cisco, Salesforce, Hugging Face, Cloudflare and Red Hat, to build an open AI security stack spanning agent harnesses, identity frameworks, secure model formats, vulnerability scanning and governance tools. NVIDIA will contribute open models, datasets and its new NOOA agent research framework to accelerate AI safety research.

The announcement reflects a broader shift in how frontier AI companies are approaching cybersecurity. Rather than framing open and closed models as competing philosophies, NVIDIA argues that both are essential, with open-weight models providing transparency, customisation, and sovereign control for defenders. As AI agents become increasingly autonomous, the company is urging policymakers to treat open AI infrastructure as a strategic defensive asset instead of a security liability. This signals that the future AI race may be defined as much by secure ecosystems as by model performance.

Startup In The Spotlight: Monk CI

Founded in 2025 by Ujjwal Prashant and Nitin Mandale, Monk CI is building an AI-powered continuous integration (CI) platform to help engineering teams accelerate software build pipelines while reducing cloud infrastructure costs. As AI coding assistants enable developers to generate more code, the startup is addressing the growing bottleneck of longer build queues and increasingly expensive testing infrastructure.

Instead of requiring engineering teams to redesign their existing CI/CD workflows, Monk CI offers a one-line migration from traditional GitHub Actions runners. Its infrastructure routes jobs to pre-warmed virtual machines to reduce build times, while AI agents analyse failed builds, summarise logs and, in some cases, automatically diagnose and resolve errors. The startup claims its managed runners can reduce build times by up to 10X and lower CI costs by 75%.

Since its launch, Monk CI has onboarded more than 15 companies, secured key industry certifications and begun pilot deployments with customers in the US. Looking ahead, the startup is betting that as AI-generated code dramatically increases the volume of software builds and tests, enterprises will increasingly require AI-native developer infrastructure that delivers faster pipelines, lower compute costs and more autonomous software delivery workflows.

Prompt Of The Week

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

Here’s the prompt Darshil Rathod, cofounder of Linkrunner, an AI mobile measurement platform, uses to turn raw customer data into concise, actionable recommendations instead of lengthy reports.

“A customer asked:

[Paste their exact question, word for word]

Here is the relevant data:

[Paste the export, query results, or campaign numbers]

Answer their question with an actionable insight, not a report.

Rules:

• Start with the direct answer in one or two sentences.

• Support every claim with a specific number from the data provided.

• Clearly distinguish what the data proves from what it cannot prove. Do not make assumptions.

• End with the single action I should recommend and the expected outcome if the customer follows it.

• Keep the response short enough to send as a WhatsApp message.”

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 The AI Layer Powering India’s Farms appeared first on Inc42 Media.