On 3 August 2026, Atlas Intelligence moved from design into implementation planning. This is the phase in which the architecture stops being a set of ideas on a whiteboard and becomes a set of decisions about what will actually be built, in what order, and against which inputs.
Scoping the first domain
The first domain implementation is Atlas Energy. Planning for it covered four areas:
- Data sources. Which operational and external feeds Atlas would need to connect to, and how each of them would be ingested, verified, and maintained over time.
- Corridor coverage. Which geographic corridors into and out of Malawi would be modelled first, and what operational behaviour in each of them would need to be understood.
- Depot models. How the network of depots, their capacity, their inventories, and the movement of product between them would be represented inside the platform.
- Intelligence outputs. What the platform would actually produce for its users: forecasts, risk indicators, early warnings, and the shape of the explanations that would accompany them.
Why this phase matters
It is tempting to move from a strong architectural idea straight to writing code. In practice, the decisions that determine whether a system like Atlas succeeds or fails are made during planning: what data is trusted, what the models assume, what the platform refuses to do when it is uncertain.
Planning is where those decisions get made explicitly, rather than by accident during implementation. We would rather spend a month answering these questions in a document than spend a year discovering the answers in production.
The decisions that determine whether a system succeeds or fails are made during planning, not during implementation.
What comes next
Implementation began shortly after this planning phase concluded. The work that followed is described in subsequent entries.