15 Nov 2026 Who gets to change the brief?
Waqar Rizvi
Community-led technology is easy to claim while everyone agrees. The harder test comes when local knowledge says the plan is wrong.
As a journalist, I have always disliked the logic of parachute journalism: arrive with the story already half-formed, gather a few local voices, file, and leave. The problem is not that outsiders report on places they do not come from. Some of the best journalism crosses borders. The problem begins when the people closest to a reality are useful mainly as sources, while somebody farther away keeps the authority to decide what that reality means.
International development has spent years trying to move away from its own version of that habit. The vocabulary has certainly improved. Projects are now “locally led”, “community-centred”, “participatory” and “context-sensitive”. Yet a local organisation can deliver the programme, employ local staff and maintain trusted relationships while the problem, budget, timetable and measures of success have already been settled somewhere else.
Technology projects can reproduce the same arrangement. A community may supply the context, data, labour and legitimacy while somebody elsewhere still defines the problem and decides which objections count. So the distinction that matters is not participation versus exclusion. It is the difference between being heard and having enough authority to change something.
A study of humanitarian localisation in Yemen, published in Development in Practice in August 2026, captures the frustration well. Researchers from the Doha Institute for Graduate Studies quote one representative of a Yemeni local non-governmental organisation: “Localisation in Yemen has evolved over time, but real authority has not necessarily shifted. Everyone uses the word, but the same structures stay in place.”1 When local knowledge says the original plan is wrong, what are local people actually allowed to change?
When local leadership stops at delivery
The Yemen study found local organisations doing much of the implementation, including in difficult and insecure settings, while important decisions over programme design, budgets and compliance remained elsewhere. Another local representative described the relationship with international organisations as being treated “as subcontractors rather than as true partners”.1 Short grant cycles, limited recovery of overhead costs, demanding reporting requirements and strict audits shifted substantial operational and compliance burdens onto local organisations while restricting investment in their own staff and systems. Domestic authorities added another layer of constraints through registrations, permits and approvals.1
The researchers call part of this pattern “risk-transfer lock-in”. The phrase is technical, but the mechanism is not. Weak access to flexible or core funding can stop a local organisation from building the systems international partners say it needs; the same capacity gap can then be cited as a reason for withholding larger, more flexible or more autonomous roles. An organisation can become highly experienced at delivering somebody else’s programme without gaining much more power over the programme itself.1
Yemen also prevents an easy North-versus-South morality tale. Local authorities can hoard power; local organisations can be hierarchical; international partners sometimes insist on safeguards that vulnerable people genuinely need. The problem is not the passport of the person holding authority. It is whether that authority can be questioned. Local staff, local offices and local-language materials matter, but they still do not tell us who can alter the brief.
Who gets to be the expert?
The imbalance can begin before a programme has a budget or beneficiary list, with a quieter question: whose knowledge is treated as knowledge? A 2022 study by Humanitarian Advisory Group, Pakistan-based GLOW Consultants, CoLAB and inSights looked at whose work humanitarian research cites. In the documents it reviewed, 95% of cited references came from Global North institutions, 3% from Global South institutions and 2% from joint North-South work. The researchers stressed that their sample was limited, but the imbalance was striking.2
The study also found Global South actors more often treated as sources of primary data than as authors, analysts or designers of research frameworks. A collaborator from Bangladesh put it plainly: “We cannot be a researcher until we get the proper recognition or authorship for the publication.”2 Being asked what you know is not the same as being trusted to decide what the knowledge means.
For me, this is where the old “West knows best” instinct can survive without anyone saying the words aloud. One kind of knowledge arrives already recognised as expertise; another arrives as “local context”. That does not make Global North knowledge inherently suspect, or Global South knowledge automatically right. The same hierarchy can be reproduced locally too. The better test is whether local knowledge can alter a consequential decision.
AI can widen capability without shifting power
That question has become harder to ignore as artificial intelligence enters development and humanitarian work. Research published in September 2026 by Pakistan-based GLOW Consultants and Humanitarian Advisory Group documents local humanitarian organisations in Pakistan using AI for reports, proposals, analysis, translation and communication. The benefits are real: these tools can make organisations faster and more capable participants in the humanitarian system.3
The report’s conclusion is blunt: artificial intelligence “does not, on its own, shift the power, resources and decision-making needed to advance localisation.”3 That is easy to miss because capability is visible. A better proposal, faster analysis or more polished report looks like progress, and often is. But an organisation can become excellent at navigating a system without gaining much say over what the system funds, which risks it accepts, how success is measured or what happens when local judgement conflicts with the original design.
Technical safeguards are part of the answer, not a substitute for transparency, contextual knowledge and accountable human judgement. But that still leaves a basic governance question. A person can review an AI output and still have no authority to change the rule behind it. A community can be consulted about a technology without having a meaningful say over how it is used. “Human in the loop” is not the same thing as community at the controls.
In practice, the governance questions are fairly ordinary. Who decided that AI was the right tool in the first place? Who can reject an output that conflicts with knowledge on the ground? Who can suspend its use when circumstances change? Who controls the data afterwards? If a technical specialist overrules a community objection, who has to explain that decision, and to whom? A system can answer every technical-safety question correctly and still leave all consequential authority somewhere else.
A village where the plan could move
Cenrana, a small village in South Sulawesi, Indonesia, offers a useful contrast. Commercial providers had little incentive to build the internet service the village needed, so its government used Indonesia’s Dana Desa, or Village Fund, to finance a local network. The proposal went through a Musyawarah Desa, a village deliberation in which residents, community leaders and the village council discuss development priorities. A local internet service provider supplied technical advice and connectivity but held no equity in the network. The village approved the plan and budget and paid the provider as a supplier.4
The arrangement did not freeze once the equipment was installed. Fallen trees damaged parts of the original network, so the configuration changed. Voucher revenue later covered operating and maintenance costs, residents hosting routers or towers earned commissions, and income was reinvested. By 2025, the network was being formalised as a village asset.4
The detail I keep returning to is the price. Villagers wanted a better deal on internet access. The International Telecommunication Union records that pricing was negotiated with the community and that the village government adjusted the voucher offer when residents asked for lower prices.4 Nothing dramatic happened, and that is exactly why the example works. People wanted a material term changed, and the people governing the service had enough authority to change it. No distant project office had to decide whether the original design could be disturbed.
Cenrana is not proof of perfect community governance. The case study tells us much more about planning, financing and network management than about who has the strongest voice inside a village meeting, or what happens if someone wants to challenge village leadership. Those questions matter because “community” is never a single person with a single interest.
Still, the case lets us follow the authority rather than admire the label. The village helped set the priority and financed the infrastructure. The outside provider supplied expertise without owning the network. The service could be adjusted when residents wanted a change. External dependence remained, but it did not swallow every important decision. That is a more realistic model of local control than expecting communities to do everything alone.
Follow the authority
Five questions help distinguish community governance from community-facing language. They work best as a chain rather than a checklist. Who sets the priorities tells us where a project begins. Who owns it tells us what remains when a donor or contractor leaves. Where decision-making authority sits tells us who can change course. External dependencies show where that authority may still be constrained. Accountability tells us whether the people holding power, international or local, can themselves be challenged.
Accountability is easiest to understand when it is concrete. In separate research on locally led humanitarian action, a community member in Pakistan said: “I could visit the office or I can raise concerns with some of the staff there … I know that the feedback process is anonymous.”5 The description is modest but useful: the person knew where to go, knew concerns could be raised and believed there was some protection in doing so.
Moving authority closer to a community does not automatically make it fair. Women, younger people, people with disabilities, poorer households, minorities and anyone dependent on local institutions may not have the same practical freedom to disagree. A locally legitimate process still has to make room for challenge. Accountability can be locally intelligible: a meeting in a familiar language, a trusted complaints route, an elected committee or customary structures. The form will vary. The principle should not: being heard has to be capable of changing something.
For funders and technology partners, that means deciding some things before a disagreement happens. Which decisions can local actors change themselves? Which require joint approval? What cannot be changed, and why? That means realistic funding for overheads and organisational capacity and agreed room to move part of a budget when needs change. It also means sharing authority over major programme changes and being clear about who controls data and other assets after a project ends. The Yemen study shows why this matters: when every meaningful adjustment requires approval elsewhere, the partnership may be local in delivery while remaining external in authority.1
Communities also need resources to participate seriously. Translation, transport, accessible meetings, childcare, independent advice and the time required to understand a technical proposal all cost something. A project that budgets for consultants, software and equipment but assumes community governance will arrive as free labour has already made a choice about whose time it values.
None of this means the local actor should always prevail. Donors remain accountable for the money they spend. Technical specialists sometimes know that a proposed change will not work. Local authorities can misuse power, and communities disagree among themselves. There will be occasions when saying no to a local request is justified.
What local actors need is not an automatic veto. They do need standing in the process: a recognised right to question a consequential decision, ask for reasons, bring forward evidence that can change it, and seek a remedy when the process fails. That asks more of a project than consultation does. It does not pretend that every participant should have equal authority over every decision.
The disagreement test
The five questions become most useful at the point of disagreement. I think of that as the disagreement test. When priorities stop lining up, is the decision explained? Can it be challenged? Can local evidence overturn an earlier assumption? Is there a route to remedy? Or was participation meaningful only while everyone agreed? That is the point at which participation becomes governance.
A project can do useful work without being community-governed, and there is nothing wrong with describing it accurately. What should make us uncomfortable is using the language of local leadership when local people are trusted with implementation but not with consequential disagreement.
Consultation tells us who was invited into the room. Disagreement tells us whether they were expected to do more than agree.
When conditions change, when local knowledge contradicts the plan, or when a community wants something the funder did not anticipate, who gets to change the brief? The answer usually tells us where the power was sitting all along.
* AI disclosure: The author conceived and developed the article, set its argument and framing, selected and interpreted the evidence, and made all substantive editorial decisions. OpenAI’s ChatGPT supported source discovery, cross-checking, and wording and revision suggestions. The author determined the final text and takes full responsibility for it.
Notes
1. Ahmed Almaweri, Hasan Alawami and Wjdan Almatari, “Why humanitarian localisation stalls in conflict settings? Administrative vetoes and risk transfer in Yemen,” Development in Practice, published online 21 August 2026. https://doi.org/10.1080/09614524.2026.2702522
2. Humanitarian Advisory Group, GLOW Consultants, CoLAB and inSights, Needles in a Haystack: An Analysis of Global South Roles in Humanitarian Knowledge Production, December 2022. https://humanitarianadvisorygroup.org/insight/needles-in-a-haystack-an-analysis-of-global-south-roles-in-humanitarian-knowledge-production/
3. Zaki Ullah and Tehreem Tassaduq, GLOW Consultants, Local Leadership in the Age of AI: How Artificial Intelligence is Transforming Locally Led Humanitarian Action, Humanitarian Advisory Group, 16 September 2026. https://humanitarianadvisorygroup.org/insight/local-leadership-in-the-age-of-ai-how-artificai-intelligence-is-transforming-locally-led-humanitarian-action/
4. International Telecommunication Union, Can the Dana Desa programme bridge the digital divide in rural Indonesia?, 2026. https://www.itu.int/pub/D-GEN-INDONESIA-2026
5. Helen Guyatt, “Unpacking the value of locally led humanitarian action,” Humanitarian Practice Network, 15 August 2022. https://odihpn.org/en/publication/unpacking-the-value-of-locally-led-humanitarian-action/
Waqar Rizvi is a Canadian-Pakistani researcher, journalist and policy consultant working internationally across technology, public-interest media and policy. His work with media, civil society, research and consulting organisations has focused on information integrity, AI ethics and governance, media intelligence and regulatory analysis. He holds an MA with Distinction in Media, Ethics and Social Change from the University of Sussex.