Project by Projects

What should philanthropy leave behind when the technology moves on?

There is a certain rhythm and flow to things that we are beginning to notice in the way tech funding moves. Especially for and around emergent tech. 

A new tool shows up, its possibilities become visible, and almost immediately the language of transformation begins to gather around it. Before long there are calls for proposals, innovation challenges, accelerator programmes, pilots, demonstrations and a growing sense that this particular technology might finally help us solve some long-term sticky problems that have been resistant to other approaches taken.

It is hard to ignore the stories, like one about a health worker supported by an AI assistant. Or about a farmer receiving advice through a local-language platform. You’ve probably read about the charity automating part of an already overburdened case-management process. A government department testing predictive and/or civic-tech tools. A start-up promising to reach people more quickly, more cheaply and at a scale that existing systems have struggled to achieve. And sometimes, importantly, these interventions do work. They save time. They expand access. They create possibilities that were not there before.

We write this with some caution because not as technologists, nor are we pretending to sit at the centre of every one of these experiments. This writing is from the position we know best: watching the field closely, reading the research and policy conversations as they develop, and learning from colleagues who are much deeper in the work — philanthropists trying to decide what to fund, technology founders building these systems, and global development practitioners trying to work out how emergent technology can genuinely serve the public good. 

What we are increasingly interested in is not simply whether the technology works, but what happens within the system around it, when it does. The lingering question ….. what remains when the project ends, the grant closes and the technology changes?

Because technology moves quickly, but institutions rarely do. Relationships take time to build. Public trust takes even longer. The capacity to govern a technology, maintain it, question it, adapt it, replace it or decide that it should no longer be used is not created simply because the technology has been introduced. In many cases, those capabilities have to be built deliberately, and often they are the least visible part of the intervention.

The real test of “AI for Good” cannot only be whether a technology delivers something useful today, but whether that intervention leaves something valuable behind tomorrow. Would it help create stronger institutions, better judgement, deeper local capability, more resilient systems, clearer rules, better public infrastructure, or even just a community that is better positioned to shape what comes next.

That right there, is where the conversation about technology for good starts to become much more interesting. That is what inspired this piece you are reading today and the snapshot of case studies that we find most interesting currently. 

🧭 The shift philanthropy needs to make

Much of the public-interest technology conversation begins with the intervention:

• What tool are we building?
• What problem will it solve?
• How many people will it reach?
• How quickly can it scale?

These are smart and definitely reasonable questions to ask, but it seems, in asking these questions, we tend to forget that tech intervention does not walk into an empty space. It is often brought into an ecosystem of organisations, public institutions, rules, infrastructure, communities, workers, funders and existing systems. 

Paying attention to that means asking hopefully smarter and more resonant questions (and if we are lucky these inform the way the impact of the tech is measured);

• What happens to all of those other actors when we introduce something new? 
• Does the intervention strengthen what is already there, or quietly work around it? 
• Does it build capability within a public institution, or create a parallel capability that disappears when the funding does? 
• Does it help a local organisation do its work better, or gradually make that organisation dependent on infrastructure, expertise or a vendor it cannot afford without continued external support? 
• Does it create knowledge that others in the ecosystem can use, or does the learning remain inside one funded project?

If philanthropy is serious about technology in the public interest, I suspect this is where some of the most important learning will be found. The ambition should surely not only be to fund a better intervention. It should be to understand whether, through that intervention, we have helped leave behind a stronger ecosystem for everyone who has to live with, govern and build upon the technology afterwards.

Now let’s see a few case studies that demonstrate this in practice:

🌱 Case 1: Funding tools…. or funding the conditions around them?

In 2025, the Gates Foundation committed more than US$7.3 million to support the Rwanda AI Scaling Hub. The hub is intended to scale inclusive AI solutions in maternal health, agriculture and education through partnerships, ethical governance and enabling infrastructure.

This is significant because the investment is not framed solely around one product. It recognises that AI solutions require a surrounding system:

• partnerships;
• governance;
• infrastructure;
• coordination;
• and pathways to scale.

A separate Gates Foundation grant of more than US$4.6 million supports a Nigerian platform intended to scale responsible, gender-intentional AI in health, education and agriculture through inclusive partnerships for public benefit.

These grants are a welcome pointer to an important evolution in philanthropic thinking. The unit of investment is beginning to move from a tool towards a national or sectoral capability.

That is promising, but yet even platform and hub approaches raise difficult questions.

Questions global development actors should ask

🟡 Who governs the hub?

🟡 Who determines which problems deserve attention?

🟡 What happens when donor funding ends?

🟡 Does the hub strengthen public institutions—or become a parallel institution?

🟡 Are local universities, regulators and civil-society organisations gaining lasting capability?

🟡 Who owns the data, models and intellectual property developed through the programme?

A hub can coordinate an ecosystem, but it can also centralise influence. 

A platform can create shared infrastructure, while also becoming another gatekeeper.

So how do you create lasting-good in this situation?  You ensure that in tracking and measuring impact and influence, you ask questions beyond scale. Ask – does this hub distribute capability?

🏗️ Case 2: From isolated innovation hubs to public innovation systems

In April 2026, Nigeria launched a nationwide network of University Innovation Pods, developed through a partnership between the federal government and UNDP.

UNDP described the initiative as the first national scaling of its UniPod model through direct government co-investment—an explicit move from pilots towards a nationally owned innovation system.

Ethiopia has also established an AI UniPod through the Ethiopian Artificial Intelligence Institute, UNDP and Addis Ababa University. The facility includes high-performance computing, robotics equipment, laboratories and collaborative research spaces.

This is closer to the idea of lasting good. Universities can become enduring sites of:

• technical training;
• research;
• experimentation;
• public-sector problem-solving;
• and collaboration between government, academia and industry.

Yet we know that infrastructure is not the same as an institution. For instance, a laboratory can be opened without a long-term maintenance budget or equipment can be purchased without building a sufficiently deep faculty.

The thing is that as we learned from implementing civic-tech innovations previously, innovation challenges can produce prototypes without creating procurement pathways through which public institutions can adopt them. Just as a situation where young people can be trained without having access to jobs, capital, research funding or computing resources afterwards.

Remove two or three of these elements and the result may still look good and will be celebrated. BUT it may not endure. The opportunity for philanthropy and global development that is intentional, is to go one step beyond supporting innovation spaces to connecting them to sustainable systems:

• public universities;
• government research priorities;
• regional academic networks;
• local capital;
• public procurement;
• and independent evaluation.

⚕️ Case 3: When evidence becomes the intervention

One of the more promising recent philanthropic developments is the Evidence for AI in Health initiative. In February 2026, the Gates Foundation, Novo Nordisk Foundation and Wellcome announced a joint investment of US$60 million to support locally led evaluations of AI tools in primary and community healthcare in lower- and middle-income countries.

The purpose, beyond developing more health tools, is to help governments and health systems determine:

• which tools work;
• under what conditions;
• where they create value;
• and how they can be used responsibly.

This is great because the development sector has often treated evidence as something produced after an intervention to prove that it succeeded.

But with emergent technology, evidence must do more. It should help institutions decide what not to adopt. It needs to expose where performance declines across languages, regions, genders or population groups. And it should identify hidden labour and implementation costs.

You could also add that it should clarify whether the technology improves outcomes or merely moves work from one part of the system to another. And if you are asking that, then one must ensure that it leaves governments and local research institutions better able to conduct future evaluations themselves.

The important distinction

1. Evaluation FOR the project

 versus

2. Evaluation CAPABILITY for the system

The first tells us whether one intervention worked. The second strengthens the system’s capacity to assess every intervention that follows.

💻 Case 4: Compute access can unlock innovation …… but what happens afterwards?

At the Nairobi AI Forum in February 2026, a partnership involving the AI Hub for Sustainable Development and Cineca announced 1.5 million GPU hours for 130 African innovators.

The support is intended to advance work in areas including food security, climate resilience and local-language voice technology. The wider initiative also anticipates investments in data infrastructure, regional data arrangements, technical support and pathways from proof of concept to finance.

This is laudable. Compute is one of the largest barriers facing African researchers, non-profit organisations and early-stage technology companies. Without it, they cannot train, test or adapt more advanced systems.

But philanthropic and development actors must be careful not to confuse temporary access with lasting capability.

Think about it, a researcher may receive computing credits for six months, but then what happens in month seven? Or perhaps a start-up may develop its tool on externally controlled infrastructure, BUT can it migrate later?

Philanthropy and global dev actors may need to distinguish between:

• subsidising access;
• creating shared access;
• and building institutions that govern access.

These are not the same thing.

🌍 What local charities and civil-society organisations may be missing

Large foundations and development agencies are not the only actors adopting emergent technology. Local charities and non-profit organisations are increasingly using AI to:

• draft proposals;
• analyse survey data;
• generate communications;
• translate materials;
• automate donor engagement;
• manage beneficiary records;
• produce reports;
• identify patterns in programme data;
• and reduce administrative pressure on small teams.

Many of these uses are sensible and may help overstretched organisations do more with limited resources. However, the wider effects are not always visible.

A charity may save time while simultaneously:

⚠️ uploading sensitive beneficiary information into an external system;

⚠️ losing internal writing or analytical capability;

⚠️ introducing errors that staff do not have time to verify;

⚠️ standardising community stories into donor-friendly language;

⚠️ relying on a free tool that later becomes unaffordable;

⚠️ weakening relationships with local translators, researchers or creatives;

⚠️ or allowing an algorithm to influence decisions that were previously based on human judgement and local knowledge.

This does not mean charities should avoid AI.

It simply means adoption should be understood as an organisational and ecosystem decision—not merely a productivity decision.

🌊 Every intervention creates ripples

The case studies above point to something I think we need to hold more deliberately: technology interventions rarely stay within the boundaries of the project that funds them. A platform may reach 100,000 farmers and improve access to advice, but it can also change the role of extension workers, shift control over agricultural data, create new dependencies or exclude those without connectivity, devices or dominant-language literacy. Some of those effects may be positive. Some may not. But all of them are part of the impact story.

This is where we need to move from asking only whether a project worked to asking what it did to the wider system. Did it strengthen institutions or bypass them? Build local capability or deepen external dependence? Create shared assets or concentrate value elsewhere? The intervention can succeed at one level and still create fragility at another.

That is the thinking behind PxP’s emerging hypothesis for “lasting good”: a simple way of looking beyond immediate outputs to ask whether technology investments leave people, institutions and ecosystems more capable than they were before.

🧩 The PxP Lasting Good Framework: From Project Impact to Ecosystem Impact

Our hypothesis at PxP is fairly simple: emergent technology requires us to widen what we mean by impact. Philanthropy has become relatively good at measuring project outputs — what was built, launched or delivered — and increasingly better at measuring outcomes — what changed for the people or institutions the intervention was intended to serve. But technology also changes the systems around it. That gives us a third layer to consider: ecosystem impact.

Level	The question	What we might look for
1. Project Output	What was created or delivered?	A platform launched, a model trained, people reached, services deployed.
2. User / Institutional Outcome	What changed for those directly involved?	Faster services, improved diagnosis, time saved, better access to information.
3. Ecosystem Impact	What changed in the wider system because this intervention existed?	New capability, stronger or weaker institutions, shared infrastructure, new dependencies, shifts in power, trust, access or inequality.

To make that third and important layer most practical, we have been experimenting with seven questions that funders, development organisations and their partners might ask before, during and after an emergent-technology investment:

1. Does it solve a real public problem?
Not simply can AI be used here?, but is technology actually the right response? Sometimes the better investment may be people, processes, records, relationships or stronger institutions.

2. What capability remains?
When the grant ends, who knows more? Who can operate independently, govern the system, adapt it, maintain it or replace it?

3. What shared or public asset is created?
Does the investment leave behind something others can build upon — a dataset, local-language resource, trained institution, public standard, reusable tool, evaluation method or shared infrastructure?

4. Who holds power?
Who ultimately controls the technology, data, infrastructure and intellectual property? Who determines the terms of access? And importantly, who has the ability to leave or change providers?

5. What happens to the wider ecosystem?
Does the intervention strengthen existing local actors or bypass them? Encourage collaboration or intensify competition for scarce resources? Build shared capacity or create yet another standalone system?

6. What happens after the grant?
Who pays for maintenance, security, staff, storage, model updates, audits and user support once the initial funding disappears?

7. How will we know what we did not anticipate?
What mechanisms allow communities, workers and institutions to provide feedback, contest decisions, report harm and change course when consequences emerge that the original project design did not foresee?

None of these questions are intended to make emergent-technology funding impossibly complicated. Quite the opposite. They are an attempt to make visible some of the choices that are already being made, whether we acknowledge them or not. If we are serious about moving from AI for Good to lasting good, perhaps success needs to mean more than a useful tool or a successful pilot. It should also mean leaving behind greater capability, stronger institutions and an ecosystem better equipped to shape whatever comes next.

🎯 What philanthropy might fund differently

This could mean funding:

🔵 public-interest compute for universities and civic organisations;

🔵 local-language datasets governed with communities;

🔵 independent technology-assessment institutions;

🔵 procurement capability inside governments;

🔵 open standards and interoperable systems;

🔵 civil-society capacity to scrutinise automated decisions;

🔵 long-term maintenance rather than only product development;

🔵 regional research and regulatory networks;

🔵 locally led evaluations that inform public adoption and

🔵 institutional learning about when technology should not be used.

From AI for Good to AI that leaves good behind

The next phase of public-interest tech and civic-tech must see clearly that a successful intervention should not only leave behind a tool. It should leave behind stronger people, stronger institutions, shared public assets and a system that is better able to govern whatever technology comes next.

Perhaps the real measure of AI for Good is not simply whether the technology creates a positive outcome today, but whether it leaves behind stronger people, stronger institutions, shared public assets and an ecosystem more capable of shaping whatever comes next.

At Project by Projects (PxP), we continue to learn alongside the field, and we welcome conversations with philanthropies, development organisations and public-interest actors looking for thought partnership or practical support in thinking through their AI and tech-for-good portfolios, or in architecting the backbone systems and support that the wider impact ecosystem will need to thrive.