ON MAY 16, 2026, the São Paulo bookshop of Expressão Popular, the publishing house of Brazil's Landless Workers' Movement (MST), was packed with farmers, agroecology technicians and movement organisers. They had gathered for the launch of IARAA, an artificial intelligence platform for agrarian reform and agroecology. The tool was not built by a technology corporation. Its knowledge base, spanning movement literature from 1964 to 2026, was curated collectively by MST and World March of Women cadres; its prompts and workflows were designed by land reform organisers and community educators working under the guidance of just two programmers. IARAA will not recommend pesticides, because the movements that built it chose a knowledge base grounded in agroecological practice. It runs primarily on open-source models, on infrastructure the movement controls, and it cites its sources in every answer. Carolina Cruz of MST's technology front put it simply: "the farmers are inside IARAA, developing the platform." Such deployments are not limited to Brazil. Already, adapted versions are being tested by cocoa farmers in Africa.

This episode should provoke serious reflection within our own movement.

In an earlier article in these pages ("Whose Intelligence? AI in the Grip of Monopoly Capital"), this author documented how generative AI under monopoly capital serves speculation, surveillance, job destruction and the imperial war machine. That critique stands. But it is only half the analysis, and a half-truth frozen into an attitude becomes an error. Across the party, in mass organizations and in the broader progressive movement, AI is discussed almost exclusively as a threat, when it is discussed at all. Even though there are notable exceptions such as the work on Telugu Large Language Model (LLM) in Swecha, many comrades have concluded, understandably but wrongly, that the technology itself is the enemy. This is the position Marx and Engels criticised in the machine-breakers (luddites) of the nineteenth century: the workers who "direct their attacks not against the bourgeois conditions of production, but against the instruments of production themselves". AI is a productive force of enormous power. The question is never whether a productive force will be used, but by whom and for what. The ruling classes have answered for themselves, with over $650 billion in infrastructure spending by four US corporations in 2026 alone. We have not yet answered for ourselves.

Lenin insisted that a communist must "enrich his mind with a knowledge of all the treasures created by mankind". The Marxist method of totality demands that we grasp economy, politics, culture and ideology together. Yet the volume of human knowledge is now too vast, too fast-changing and too specialised for any cadre, least of all one carrying front-line organisational responsibilities, to master by traditional means. Ironically, the scholars serving capital have the databases, the research assistants and the leisure; our comrades have none of these. It is precisely here that AI, wielded consciously, can break the elite monopoly on knowledge production. Movements in the Global South are already demonstrating this in practice.

Consider the labour of research itself. A serious political-economic investigation may require reviewing hundreds of web pages, working through hundreds of downloaded documents and extracting from them scores of pieces of hard evidence. Work of this order once consumed months of a research team's time; with AI assistance it can be completed in hours. A party research cell preparing a note on, say, the fiscal record of a state government or the ownership structure of the Adani group can now work from all the available material, not a thin sample of it.

The same holds for analysis. Hundreds of hours of field interviews can be transcribed and placed in a dedicated database that a researcher can interrogate exhaustively; one researcher discovered information in his own recordings that he had not registered during the interviews themselves. For a party that lives by concrete investigation of concrete conditions, this is the mass line supercharged. Nor is specialised method a barrier any longer: the accumulated techniques of entire disciplines, from statistical analysis to argument mapping, can be invoked in plain language, without programmers as gatekeepers.

The gains extend to everything our movement publishes. A weekly newsletter, of the kind many organisations have abandoned because it devoured cadre labour, can be substantially automated: AI gathers a wide pool of news leads, human editors select the stories worth telling, AI then researches each story from multiple angles across hundreds of pages and drafts summaries in the movement's own style for final human editing. Pamphlets, leaflets, videos, social media content and study materials in multiple Indian languages are all within reach of the same approach, at a fraction of today's cost in cadre time.

Comrades will object: do these models not hallucinate, and do they not carry the ideological imprint of the Western internet they were trained on? They do, and the objection deserves a serious answer rather than dismissal. The answer is that these defects can be countered by design, not by abstinence. Typing questions into a web chat interface such as chatgpt.com or claude.ai will indeed reproduce every one of these problems: a model left to its defaults repeats imperialist, neoliberal assumptions dressed up as neutrality, and invents facts with fluent confidence. But open-source agentic frameworks can be carefully assembled to work differently. The researcher declares the analytical standpoint explicitly, supplying Marxist political economy frameworks as the foundation before any analysis begins; the model is grounded in curated document bases rather than the open internet; and verification engines are built into every step of the workflow, checking claims against sources before anything is published. IARAA embodies this discipline: it is instructed to cite sources in every response and to refuse to answer when its knowledge base cannot support one.

Nor must our movements chain themselves to any single corporation. In practice a mix of models, OpenAI's GPT, Claude, DeepSeek, Qwen, Kimi and GLM among them, can be deployed with an understanding of which model suits which task, relying predominantly on the Chinese models to keep costs down, for they are fast catching up with the capabilities of the US frontier models at a fraction of the price: Open Source Models such as Kimi 3, GLM 5.2 and  DeepSeek V4 cost an order of magnitude less than frontier US Models such as Claude or ChatGPT with near equal capabilities. When Oxfam reports that three billionaires control nearly 90 per cent of the global chatbot market, the political meaning of this plurality is obvious.

Nor does any of this displace the human core of political work. Practitioners have found that when AI is asked to make the final editorial selection for a movement publication, the product immediately turns mediocre; knowing what to say to comrades is a political and human judgement no machine can make. The purpose of these tools is to strip away the mechanical desk labour so that cadres spend more time in the villages, factories and bastis, with the masses, completing the loop between theory and practice.

Socialist China shows what this looks like as state policy. On April 10, 2026, China's Ministry of Education released an action plan to make AI ubiquitous across every level of education by 2030: AI literacy built into teacher certification, automated grading to reduce teacher workload, learning companions for students in rural and remote schools, intelligent research platforms in the universities, all resting on a national public computing infrastructure. While Indian policy debates remain trapped between corporate hype and moral panic, a socialist state is systematically equipping its teachers, students and researchers with the new productive force.

Our movement has been here before. The printing press was the instrument of colonial administration and missionary propaganda; the working-class movement seized it and built its own press, of which this newspaper is a descendant. The task now is the same, and it cannot be discharged with a resolution or a seminar. It requires practice: hands-on workshops for party research staff, pilot AI-assisted production lines in our publications and social media teams, training courses in the mass organisations, study circles where comrades learn by doing, and exchange with the movements, from São Paulo to Johannesburg to Accra, that are already ahead of us. The tools are largely open source and the running costs manageable for movements. There is no excuse for delay.

The monopolies' abuse of AI is not a verdict on the technology. It is a verdict on the class that owns it. Our answer cannot be to stand aside while the ruling class arms itself with a new productive force. It must be to master that force ourselves, and to place it, as the MST's farmers have done, in the hands of the people.