🌍
Field Applications
5Featured
Global Mapping of AI in Community Health Worker Programs
ICTworks · Feb 12, 2026
A new mapping by Nate Miller identified 38 AI systems supporting community health workers across LMICs, with 87% concentrated in Sub-Saharan Africa and South Asia. Only 34% show signs of scaling beyond pilot phase — a pattern the author calls "pilot purgatory." Over 40% of implementations are LLM-powered chatbots acting as always-on supervisors, but research shows CHWs often perceive AI as infallible, forming incorrect mental models about how these systems work.
Key Insight
Directly relevant to IDinsight's health systems portfolio. The finding that programs are launched without scaling plans, sustainable financing, or health system integration mirrors challenges we see in M&E of digital health interventions. The CHW trust gap — treating AI outputs as infallible — is a critical design and evaluation consideration.
2
IDinsight Global AI Adoption Survey Results
IDinsight Internal · #ai_for_idi · Feb 3, 2026
Shared by @Gayatri Dewan
Nearly all IDinsight staff now use AI at least weekly, with half using it daily. Across 147 respondents, average weekly time saving is 6 hours (~15% of work time). Top use cases include writing/editing, coding, research, and email drafting. The thread surfaced creative applications including AI-as-coach for writing, dictation-to-message workflows, and Claude Projects for persistent project context.
3
Where Should International NGOs Start a Generative AI Journey?
ICTworks · Feb 3, 2026
Practical roadmap drawn from Oxfam Denmark's experience. Notes most INGOs are still in the "experimentation" phase, with useful parallels to mobile/digital adoption circa 2012 — but the learning cycle is now 3-6 months rather than 2-3 years.
4
World Bank Champions "Small AI" at India AI Impact Summit
World Bank · Feb 17–20, 2026
The World Bank is centering "small AI" — practical, affordable solutions that run on everyday devices — as the path forward for LMICs. Less than 1% of ChatGPT usage comes from low-income countries, underscoring the need for approaches that work within infrastructure constraints.
5
Anthropic Partners with UK Government for GOV.UK Services
Anthropic · Jan 27, 2026
Claude integrated into UK government digital services — a model for how LLMs can be embedded in public service delivery at national scale. Relevant precedent for governments in IDinsight's operating contexts considering AI in citizen-facing services.
💬
Debates
5Featured
Where Is Generative AI Supporting MERL Improvement in 2026?
ICTworks / MERL Tech · Jan 20, 2026
Linda Raftree's comprehensive state-of-play for GenAI and MERL. Organizations that effectively use AI for MERL tend to be those with in-house data science capacity and clean data foundations. She identifies four archetypes of AI adoption: the Curious, the Doers, the Dreamers, and the Skeptics. More organizations are moving from talking about "potential AI applications" to sharing "actual AI applications," though many remain in the Curious stage.
Key Insight
The four archetypes framework is a useful lens for how IDinsight advises clients on AI readiness. The emphasis on data hygiene and internal capacity as prerequisites for AI-enabled MERL reinforces our own experience that AI amplifies existing data maturity rather than substituting for it.
2
Bias In, Bias Out? Understanding Gender Bias in LLMs
MERL Tech · Feb 9, 2026
New analysis on how gender bias manifests in LLM outputs, particularly relevant for development organizations working on gender equity programs and using AI tools for research and analysis in gendered contexts.
3
Made in Africa AI in MERL: Research Digests 1 & 2
MERL Tech · Jan 21 & Feb 10, 2026
Landscape study findings on AI adoption by African MERL practitioners. Highlights multi-layered barriers — infrastructure gaps, lagging digitalization, misaligned frameworks — alongside growing consensus that AI for African MERL must be rooted in locally defined principles and African language representation.
4
Digital Sovereignty Won't Save Us From Internet Shutdowns
ICTworks · Feb 10, 2026
Iran's internet traffic has dropped to effectively zero since January 8, 2026. The piece argues that digital sovereignty infrastructure can become a tool for authoritarian control — relevant to assumptions about connectivity and data access underlying AI deployments.
5
OpenAI Disbands Mission Alignment Team
TechCrunch · Feb 11, 2026
OpenAI's second alignment-focused team has been disbanded, with former lead Josh Achiam reassigned to a new "chief futurist" role. This follows the 2024 disbanding of the superalignment team, raising ongoing questions about safety commitment at leading AI labs.
🛠️
Tooling
5Featured
Claude Opus 4.6 Released
Anthropic · Feb 5, 2026
Anthropic's most powerful model to date, with major improvements in agentic coding, computer use, tool use, search, and financial research. Bloomberg reports the model can analyze company data, regulatory filings, and market information to produce detailed analyses that would normally take days. Also improved at creating spreadsheets and presentations.
Key Insight
The improvements in spreadsheet/presentation creation and multi-step research workflows are directly relevant to IDinsight's data analysis and reporting workstreams. The financial research capabilities could also be useful for cost-effectiveness analyses and budget modeling tasks.
2
OpenAI GPT-5.3-Codex
OpenAI · Feb 5, 2026
New agentic coding model combining frontier coding performance with reasoning capabilities, 25% faster than its predecessor. Capable of long-running tasks involving research, tool use, and complex execution. Notably the first model that was "instrumental in creating itself."
3
Gemini 3 Deep Think Upgrade
Google DeepMind · Feb 12, 2026
Major upgrade to Google's specialized reasoning mode for science, research, and engineering. Identified a logical flaw in a peer-reviewed math paper that passed human review. Now available to researchers via the Gemini API for the first time.
4
Gemma 3n: Mobile-First Open Model
Google DeepMind · Feb 11, 2026
New efficient open model (5B/8B parameters) running with just 2-3GB memory on mobile devices via a "Per-Layer Embeddings" innovation. Responds ~1.5x faster on mobile — highly relevant for low-resource field settings where connectivity and compute are constrained.
5
MERL Tech Launches UX Design of AI Learning Group
MERL Tech · Feb 3, 2026
New practitioner learning group focused on UX design for AI tools in development and humanitarian contexts. Will explore user patterns, informed consent, safety risks, and responsible design conventions. Open to UX researchers, conversation designers, product managers, and MERL professionals.