There Is Nobody Left to Replace

The debate about artificial intelligence often starts with the jobs it might replace. That concern deserves attention. But it is an incomplete starting point for agricultural advisory, where access to professional time can be the binding constraint.

GFRAS’s historical Zimbabwe profile cites roughly one extension worker for every 3,000 farmers ↗. The profile contains a 2009 staffing table and references from the early 2010s. It is evidence of a longstanding access problem, not a reliable count of staffing today. We should not turn an old estimate into a current national headline.

More recent research describes gaps between extension training and farmers’ needs around markets and emerging technologies ↗. Zimbabwe’s 2026–2030 National AI Strategy creates a policy context for locally useful applications ↗. Neither source proves that any particular AI product will solve those gaps.

The practical question for SAS is therefore smaller: can an agent reduce repetitive information work while preserving the adviser’s time for judgement, field visits and relationships?

The proposed behaviour is retrieve, clarify or escalate. A routine answer must come from approved material that matches the crop, region and stage. An unclear request can trigger a small number of questions. A diagnostic request, an uncertain match or a high-risk chemical query goes to a named human.

Retrieval reduces the scope for invention; it does not guarantee correctness. A document can be outdated, a passage can be misapplied, and a translation can change the meaning. Source references, validation and human review are part of the product.

This also fits SAS’s wider approach to business automation: connect agents to an organisation’s real knowledge and workflows, define what they may do, and keep consequential decisions with an accountable person.

Zimbabwean digital agriculture did not begin with generative AI. Welthungerhilfe documents Kurima Mari’s launch in 2016 and its focus on agricultural information and market access, alongside AgriShare’s equipment-sharing role ↗. These services deserve study, not dismissal.

The landscape has also moved on. Econet publicly describes EcoFarmer AI as a source of agricultural insights and recommendations ↗. That makes a blanket claim that no incumbent addresses interactive advice untenable. Provider descriptions still need independent testing before performance comparisons are made.

SAS’s proposed contribution is a client-specific knowledge and escalation workflow. Whether it offers an advantage must be tested against existing services, including the quality of handovers, local-language understanding and the time it saves.

NewsDay’s July 2026 account of the 2025 ZIMSTAT/POTRAZ household survey reports household internet access of 75.5% ↗. This review located that reporting, rather than independently inspecting the final survey report. Household internet access does not establish rural WhatsApp use, smartphone ownership by every farmer, or dependable service in a target ward.

A deployment needs local evidence about handsets, affordability, signal, charging and language. WhatsApp is a channel candidate. SMS, voice or USSD may be alternatives where commercially and technically available. None should be promised before access and operating costs are confirmed.

This article presents an argument, not a pilot result. No field results are claimed for the architecture described here. Evidence from other digital-advisory programmes is informative, but it cannot establish how this design will work in Zimbabwe ↗.

The next test is a bounded deployment with an approved knowledge base, an available human escalation team and a baseline time audit. Measure answer errors, escalation quality, repeat use, response times and cost as well as time saved.

Job impacts should remain part of the discussion. But the immediate opportunity is to make expertise more accessible and operations more effective. An agent earns its place when it helps a real team serve more people, with evidence and accountability. That is the claim SAS should make, and the one we should be prepared to test.