African developers have until October 15, 2026 to enter a challenge that asks them to build open-source AI agents for smallholder farming, and the deadline is days away. The agriculture track, titled “The Extension Backlog,” is one of four tracks in a competition that its organisers call the African Agentic AI Design Challenge (the site also uses the name Agentic AI for Africa Innovation Challenge). It is run by the Nairobi AI Community and Qhala Trust. The other tracks cover healthcare, governance and education. Entry is open to developers aged 18 and over who live anywhere in Africa, alone or in teams of up to three. You can apply through the agriculture track submission form, and the official challenge site lists all four track briefs and the rules.
The brief is more specific, and stricter, than most coverage suggests. Teams must build an agent that turns scattered rainfall, price and plot data into a recommendation that a small local extension office could use, with an officer in the loop. Participants choose one of four focus areas: climate-smart advisory, market access and price intelligence, input and credit risk, or post-harvest tracking and traceability. The central rule is that the agent only drafts advice. It may never send an advisory to a farmer, place an input order or request a loan on anyone’s behalf, and a named officer must approve anything before it reaches a household by SMS, USSD or home visit. Every recommendation must cite its rainfall, price or plot source, and the page says an unsourced recommendation is worse than none. That is firmer than the “human-in-the-loop for irreversible actions” wording used in some write-ups, which would allow more automation.
The technical requirements are demanding for a seven-day build. Teams must write their own server using the Model Context Protocol (MCP), the open standard that lets AI agents connect to tools, with at least three tools, one of which takes an action such as drafting an advisory or scheduling a follow-up. They must also use one MCP server they did not write, orchestrate with open-source software such as LangGraph, CrewAI or n8n, run one complete task on an open-weights model because household data may need to stay in-country, and log every action with inputs, outputs, timestamps and the approving person. The submission needs a public repository with an open-source licence, a demo video under three minutes showing an unedited agent run, a short description, a one-page architecture note and an evaluation file with at least eight test tasks, including one failure left unfixed. Chatbot wrappers, servers that merely proxy one endpoint and anything built on real household records without consent are disqualified. Judging is out of 100 points, with the heaviest weight on agentic depth and MCP craft (30), then open-source rigour (20) and fit to the farming cycle (20). Finalists face a 20-minute live review where judges add a new requirement on the spot, and AI coding tools are allowed as long as you can explain the code.
The rewards are real but partly unannounced. The top 40 teams receive mentorship and build resources, and the top 10 get all expenses paid to the MCP Dev Summit in Nairobi, which the Linux Foundation lists for November 19 and 20, 2026. Cash prizes are planned for first, second and third place and for a Best Women-led solution, but the amounts are still marked “to be confirmed.” The judging panel has not been named, and the submission period dates are also unconfirmed on the track page. The challenge’s main page says “seven days to build” in one place and “five days” in another description, so confirm the window with the organisers at info@qhala.com or arthur.adinda@nairobiai.org before you plan around it. The page shows logos for the Model Context Protocol, Africa AI Foundation, IBM and Microsoft but does not say what role each plays, so do not assume any of them funds the prizes.
Some context in the original announcement needs sorting. It mentions the Innovate Africa Challenge from Smart Africa and the Food and Agriculture Organization (FAO) as a similar effort, but that is a different kind of competition: a proven-solution contest with a single $50,000 award to support deployment, and applications closed on August 31, 2026. The announcement also cites a working example, UlangiziAI, a WhatsApp chatbot from Opportunity International, built with Gooey.AI, that answers Malawian farmers’ questions in English and Chichewa using the Ministry of Agriculture’s production manual. NVIDIA reported more than 4,000 questions in a three-month pilot, though the 2024 article gives no user count. It is a useful reference, and also a contrast, because this challenge would reject a chatbot wrapper without autonomous tool use. We did not check the announcement’s farm-sector figures, 60% of Sub-Saharan Africa’s labour force and about 17% of GDP, nor its claim that the continent loses billions yearly to poor information, and the brief does not require a distributed ledger for traceability, although the announcement suggests one.
For African developers, the practical question is whether a week is enough, and the honest answer is that it favours people who already know MCP and one agent framework. Anyone starting from scratch should pick a single focus area, use open rainfall and price data such as CHIRPS or FAO price bulletins, build one narrow workflow end to end and spend real time on the evaluation file, since reliability and honest failure reporting carry marks. For the wider ecosystem, the more interesting signal is the design philosophy: an agent that drafts, cites and waits for a human, running on open models with logged actions, is a template that agricultural ministries and cooperatives could adopt without trusting a black box. Whether any entry survives contact with real extension offices is something the challenge’s own live reviews and, later, field tests will have to show.
Sources: Agentic AI for Africa Innovation Challenge, official site; Agriculture track brief, “The Extension Backlog”; fundsforNGOs listing; MCP Dev Summit Nairobi, Linux Foundation; FAO, Innovate Africa Challenge 2026; NVIDIA on UlangiziAI.
