Welcome back to the Frontier Tech Hub's AI mini-series. If you missed our first issues, I would recommend catching up here:
So far we’ve explored what AI actually looks like in development, how to harness it responsibly. Today we ask: are we about to waste it?
In this issue: a quicker Ebola response, the government that banned innovation and an invitation to make use of the AI tools we’ve been creating for people like you.
When the currencies fluctuate
In 2014, Ebola moved through Sierra Leone faster than the response to it. By the time the world moved, the outbreak was already out of control and 11,325 people lost their lives in the years that followed.
The cost of time was later modelled. Researchers estimated that interventions beginning just 4 weeks earlier would have averted around 8,835 deaths in Sierra Leone alone, and saved up to $203m in funding.
The current outbreak is still ongoing. And some of what made it harder to catch than it should have been isn't a mystery. Aid cuts disbanded the wildlife-monitoring teams tracking bat reservoirs in Ituri and western Uganda within days of last year's USAID dismantling. A cross-border surveillance network between Uganda and the DRC was halted entirely. As Denis Mukwege, the Congolese gynaecologist and 2018 Nobel Peace Prize winner, put it: "when health systems are deprived of funding, the spread of disease becomes inevitable."
Most of what makes a response slow isn’t simple to fix, but at least some delay is mechanical. Useful answers usually exist, but they’re sitting in thousands of evaluation reports which have been commissioned, paid for and occasionally read by someone with enough time to file through the archives.
Just two months ago, a new tool, EvalExplorer was tested under real pressure during the 2026 Ebola rapid response. The FCDO Evaluation Unit used EvalExplorer to identify which past FCDO investments had been most successful in tackling Ebola, work that would previously have taken weeks. They did it in two days. That turnaround gave the rapid response team faster, better-informed decisions to work with at a point when that mattered most.
No evidence-synthesis tool can fix funding issues or a failure of political will. But during an outbreak, time is a currency whose value rises, and any delay is measured in lives.
Come and use it yourself
EvalExplorer is one of the AI tools we’ve been creating for you: practitioners, government innovators and professionals across the international development sector who are keen to harness AI safely and usefully, in service of better, more impactful work.
OASIS* is where we’re housing them, and we invite you in. It’s free and open.
EvalExplorer turns dense evaluation PDFs into structured, searchable evidence so you can synthesise lessons across a whole portfolio.
A Transport Global Intelligence System is in development, mapping transport-sector signals globally.
DevExplorer runs AI-assisted analysis across the IATI dataset of development and humanitarian programmes, extracting insight around 1000× faster (while keeping humans in control).
And Sam, we completely agree. We hope it won’t be long either 🤞
But three tools isn’t exactly a movement. Which brings us to the reason OASIS exists at all.
The year a government banned innovation
In 2008, more than 80 mHealth applications were being piloted in Uganda. None reached national scale or shared data with each other. By 2010, the country had over 50 concurrent eHealth projects, funded by nearly as many donors, running on at least a dozen separate record systems. Community health workers were feeding data into registries that couldn't talk to one another, on behalf of funders who weren't talking to one another either.
So in 2012, the Government of Uganda called a halt: a moratorium on all eHealth pilots until unified standards could be established. Botswana did something similar, consolidating 37 record systems down to nine.
The real failure was that nobody shared what they'd learned. Every team started from scratch because there was nowhere to put what they'd built so the next person could build on top of it. The sector paid for the same thing dozens of times and got less than once.
Now look at AI in the sector today. We're independently building chatbots, drafting guidance, commissioning landscape assessments. And the organisations closest to the problems, with the smallest budgets and the fewest specialists, are the ones getting left behind.
There is another way, and it's perfectly described by one particular example. Our Adopt lead Nick shares the full story here, but the short version is this: a tool built for a local problem in Colombo ended up being adapted by the country that hosts the platform because of one norm: When you build something useful, you share it. Shared infrastructure means the next person starts where the last one finished.
That norm isn't the default in our sector but it should be. It's why OASIS is open, why the code sits in public repositories, and why we built it with IDIA rather than on our own.
What people say privately is that they're figuring out AI by themselves, inside their own organisation, slightly guessing. That's the actual condition of the sector right now, and it's the one thing a website can't fix.
👥 Alongside the OASIS platform we've launched an AI community of practice: a place where we, as development professionals, can share, discuss, and learn about the latest in AI for the development sector. Please do join us.
🎉 A quick celebration: EvalExplorer was a runner-up at the GSR Awards 2026, recognising Rachel Lineham and Ian Montago from FCDO for "Unlocking Evaluation Evidence with AI." The judges called it a standout example of how AI can be harnessed to drive better use of evidence in decision-making.
Where this leaves us: OASIS only works if it’s all of ours.
If you've built an AI tool for development practitioners, then put it on the shelf. Someone is solving your problem right now without knowing you've already solved it. If you've got a challenge that an AI tool could take on, come and talk to us about scoping and prototyping it with you. Both doors are available on OASIS.
We’ll be back in two weeks to finish the series with fresh insights from the Future of International Development in the Age of AI.
🤖 Until then, we're pleased to report that the sector has now independently commissioned its 47th AI chatbot for frontline health workers*. All 47 are excellent. None can talk to each other. Three are called AskHealthBot. We have chosen not to build a 48th, which we consider our finest contribution to the field this year.
*We’re pulling your leg… But only a bit.
In other news
This is where we share recommendations from members. Our Network is all about collective insight, so if you’ve come across something worth sharing, get in touch!
“It's almost as if we need a ready-made Standards and Assurance Framework for Ethical AI deployments,” and lo and behold, one of our Network members was involved in creating one! Check out Safe AI from the CDAC Network.
⏩ Further on the subject of tech ethics: this team share eight ethical lessons from the frontier of forensic technologies used to find missing loved ones in Mexico.
⏩ ⏩ Further on the subject of missing loved ones: in response to Venezuela’s earthquakes last month, developers used Claude Opus 4.8 and facial recognition software donated by Mexican company Lab-Co to help locate and reunite thousands of loved ones.
And finally, understand AI in 14 minutes with Anthropic’s Chloe Lubinski, who reframes the data they’re trained on: “Language is us. Language is our thoughts and our values and our fears and our wisdom. So when you train a model on language, you’re training it on us."
Got this forwarded to you? Join the FT Network here (and let us know who shared it with you so we can say thanks)