n.V6-4.04 | TWO POPULATIONS
01| "AI at IkoCiti" is too vague to design against. There are two distinct populations of agents, they serve different masters, they fail differently, and conflating them produces bad decisions.
02| Platform agents run operations. They match projects with buyers, scan for compliance problems, watch project execution for anomalies, aggregate data, keep the marketplaces moving, handle intake and scheduling. Users mostly never meet them. When they work, nobody notices. When they fail, something economic breaks quickly — a settlement stalls, a fraudulent project clears, a whole category of projects quietly stops being matched.
03| Ikosi Assistants serve registered users, and the Maveriqs are the group that matters most. This is the agent a person talks to. It knows who they are and what they are working on. When it fails, the damage is slower and worse: someone gets bad advice, acts on it, and stops trusting the platform. Trust does not come back on the same schedule it left.
04| Both populations exist at two levels. Global agents run on IkoCiti and serve the whole network. Local agents run on each CITI and serve that barrio — local intake, local scheduling, local content, the assistant that knows this Maveriq in this neighborhood. The same structure extends to White Label deployments, where a licensee runs the same architecture under their own name. Ownership does not travel with the deployment: the agents belong to IkoCiti Holding, and what they learn flows back to the centre.
05| That last point deserves its own line, because it is where a licensing business quietly loses control of its own product. A licensee who can reshape their agents freely enough eventually has a different platform wearing our name, and the data coming back is no longer comparable to anyone else's. So configuration freedom for licensees is bounded, and the boundary is a commercial decision, not a technical one.
06| What we want: an agent workforce that behaves the same way in every barrio, so that a Maveriq in one city and a Maveriq in another are getting the same quality of support and we can compare their results honestly.
07| What we sacrifice: local fit. A locally tuned agent would serve its barrio better than a standardised one. We take the standardised one, because the alternative is fifty local variants nobody can audit, compare or fix, and because comparability across CITIs is what makes the whole network learn. Local adaptation happens inside a frame set globally, and the frame does not bend for convenience.
08| One decision is not mine and I will not fake it: whether the model layer itself — which foundation models, under what terms — sits with the AI division or with infrastructure. It is an organizational question with real consequences and it is open.
