Start with the problem. Build the foundations together.

Zineb Bhaby, introduced as Norwegian Refugee Council, argues that humanitarian organisations should identify institutional problems before selecting AI. She distinguishes individual chatbot use from embedded aid-delivery systems. The latter require pipelines, expertise, maintenance and a common foundation that isolated organisations cannot readily finance. The Linux comparison is an architectural analogy, not an evaluation of humanitarian AI.

The practical mechanism is pooled infrastructure and learning, supported by multi-year commitments. This encounters an organisational tension: collective benefits require leaders to forgo some organisation-specific priorities. Funding technology as overhead and rewarding annual application demos work against maintained shared capacity.

The audience exchange substantially qualifies the keynote. Questions concern where existing initiatives can convene, whether data or application harnesses come first, and practical collaboration. Her account of the CLEAR initiative describes initial enthusiasm without continuing staff capacity: participation needed budgeted time and reporting obligations had already been set. This is a useful negative case against assuming that convening produces durable cooperation. The response on continuous evaluation links model harnesses to shared governance and cross-organisation learning.

Evidence consists of practitioner interpretation, analogies, proposals and a specific experience of stalled coordination. Costs, outcome measures, beneficiary perspectives and comparative performance are not supplied here. References to UN80, the funding reset, named initiatives and audience affiliations are recorded claims, not separately fact-checked.