Shaping the Deep Tech Revolution in Agriculture 2025

Integration of GenAI: Before and after FIGURE 2 AFTER Farmers r ely on static advisory (pr edefined messages not customized to the individual farmers) for planning their next cropping cycle; many of these advisories ar e relevant to a broad geography . When a pest outbr eak hits, the farmers do not get timely advice. Additionally , most of the advisory is one-way communication, with no follow-ups. When two- way query r esolution is critical, the farmers wait for extension agents. Sometimes they go for advice to input providers who may have a vested inter est in pr omoting specific inputs. Overall, most farm decisions ar e reactive, fragmented, time-consuming and heavily r eliant on generic support. Additionally , farm r ecor d-keeping is manual and therefore the quality of farm data that is captur ed remains poor . This af fects futur e decision-making.Farmers use a GenAI–power ed chatbot on their smartphones, which allows for two-way communication. It generates tailor ed advice for dif ferent plot sizes, soil conditions and weather patter ns in pr eferr ed languages. The farmers r eceive instant r esponses on pest outbr eaks, including possible tr eatments to manage pests. Instead of recor ding data manually , the farmers speak to the chatbot, which digitizes their data in a format that can be used by supply chain actors. It also drafts gover nment subsidy applications automatically . After GenAI integration, most decisions ar e proactive and data-driven.BEFORE Drivers and barriers to the use of GenAI in agriculture FIGURE 3 Source: Consultations with AI4AI community expertsAdvances in large language models (LLMs): The significant success of LLMs such as ChatGPT and other generative models has boosted investment in language models, including in agricultur e. Advances in natural language pr ocessing have also impr oved the usability of Gen AI in settings with low digital literacy .Data quality issues: While the availability of data for agricultur e is incr easing, accessing quality data is still challenging, especially in smallholder contexts. This decr eases the r eliability of content cr eated thr ough GenAI models. Hallucinations and errors: GenAI sometimes pr oduces information that sounds convincing but is factually incorr ect or irr elevant. Without r obust validation, farmers risk acting on false insights that may harm yields or incomes and af fect continued use. Model transferability across domains and regions: While GenAI chatbots have r eported success in certain pilots, their transferability for use in domains such as regenerative agricultur e or acr oss value chains is limited. The r eliance on LLMs also limits the availability of context-specific r ecommendations and impairs farmer trust and adoption.DRI VERS BARRIER S Increasing availability of agricultural data: The gr owing accessibility of agricultural data is providing the foundational base for GenAI in agricultur e. Additionally , the lower cost of data collection thr ough complementary technologies such as satellite-enabled remote sensing is fuelling its gr owth. Increased investments in GenAI for agriculture: Ther e has been an incr ease in investment in GenAI, with several big tech firms investing in GenAI for agricultur e. The gr owing partnership between these tech companies and local innovators and r esear ch organizations is driving the cr eation of context-specific and localized solutions that ar e better accepted in farming r egions. Shaping the Deep-Tech Revolution in Agriculture 11