For two decades, development agencies, governments and cooperatives bought technology the same way: define a system, contract its development and, at the end, receive an application. If the project "went well", the deliverable was working software.
That model is changing fast. AI agents can already write much of the code, draft reports, query databases and answer farmers' questions. When producing software stops being the hard part, the question changes: what outcome do we need, and how will we know we achieved it?
What becomes scarce
As code gets cheaper, value moves toward four capabilities no tool ships with:
- Defining the outcome. Not "a monitoring platform", but "report forest cover across 12,000 hectares every month, with source and date".
- Designing the system. What the AI does, what people do, which data is used and where the limits are.
- Verifying quality. Measuring whether answers are right, data is reliable and performance holds over time. More in Why an AI agent needs an auditor.
- Governing use. Privacy, bias, cost and compliance with donor and regulatory requirements.
Why verification matters so much
AI agents handle isolated tasks well, but their performance drops when they must sustain work over weeks or months. That's why the viable model today isn't autonomous AI, but AI with people in the loop, plus ways to audit what each agent did.
In climate, nature and agriculture this isn't new. Measurement, reporting and verification (MRV) systems exist precisely to prove that an outcome happened: emissions were reduced, a forest was conserved, a coffee lot comes from where it says it does. AI agents now need that same discipline.
What changes for your organization
Buying outcomes instead of tools means three practical shifts:
- Start by understanding. Before automating, map processes, decisions, data and people. The bottleneck is often not technological.
- Agree on indicators up front. Hours freed, response time, expert-validated accuracy, hectares monitored: what will be measured is defined before building.
- Ask for evidence, not demos. A good partner should be able to show what the system did, with which data and to what quality.
At DAP we sum this up in four steps: understand, define, orchestrate and verify. Technology is the means. The outcome, and proof that it happened, is what matters.