Manager / Senior Manager, Incubation, The AI Access Initiative

About The AI Access Initiative (2AI)

We're at an inflection point in artificial intelligence, presenting both tremendous potential opportunity and risk for people in developing countries. AI capabilities are improving on a timescale measured in months, while development systems often move on timescales measured in decades.

Much of the technology needed for positive impact exists today, but few organizations exist to bridge research, deployment, and scaled implementation: we must act now to ensure AI's benefits reach people in poverty in low- and middle-income countries (LMICs).

Incubated at Evidence Action, we're rapidly scaling The AI Access Initiative, an organization focused on scaling AI-enabled 'big bets' to benefit tens or hundreds of millions of people in poverty in LMICs. We operate at the intersection of global development actors, top AI labs, and leading researchers to drive meaningful access to AI's benefits for the 3.5 billion people living in poverty globally.

Given the scale of opportunity, we expect our portfolio to expand substantially, but to start, we're scaling two programs focused on AI in Agriculture and AI in Health across Africa and Asia.

We are led by former Evidence Action CEO Kanika Bahl, a founding member of Anthropic's Long-Term Benefit Trust, and advised by Nobel Laureate Michael Kremer; Dario Amodei, CEO of Anthropic; and Kent Walker, President, Global Affairs for Alphabet and Google. This work builds on Evidence Action's track record reaching 530M+ people with cost-effective, evidence-based programs across 9 countries in Africa and Asia, with a focus on last-mile delivery.

The Incubator

Our Agriculture and Health portfolios came out of a cross-sector analysis of near- and medium-term AI opportunities in LMICs. The Incubator is the engine that produces the next set of interventions ripe for scale. It is a new, 3-5 person research and incubation team with a single mandate: find the AI-enabled interventions that could plausibly clear a field-recognized cost-effectiveness bar at scale, test them and iterate on the delivery model, and hand the ones that survive to a program team as a fundable Big Bet.

The Incubator runs a four-stage pipeline, modeled on Evidence Action's Accelerator, which derisked programs like syphilis screening in pregnancy and in-line chlorination before they scaled: rapid screens (days), deep dives (weeks), on-the-ground pilots with light-touch monitoring and evaluation (M&E) (months), and a repeatable stack of tools, partners, and playbooks that gets easier to reuse each cycle, all aimed at producing new Big Bets.

Most ideas will not make it all the way through; the team's value comes from making clear, fast, well-reasoned calls on which to set aside and which to advance to a fundable Big Bet.

The Role

The Manager / Senior Manager, Incubation does the core analytical and field work behind 2AI's next Big Bets. Reporting to the Incubation Director, you will own a portfolio of potential programs and pieces of the pipeline from first screen to pilot handoff. For each program you own, you will run the assessment end to end and make the recommendation on what advances, what pauses, and what is set aside.

This is a builder role in a start-up organization. The screening process, the deep-dive template, and the pilot playbook do not exist yet. You will help create them, then use them, then improve them.

What success looks like in 12-18 months:

You have identified multiple AI Big Bets for 2AI to invest in at the pilot stage, each backed by an evidence review, a cost-effectiveness estimate, and a clear recommendation 2AI leadership could act on.

You have helped design on-the-ground pilots that answer decision-relevant questions.

You have built the AI-enabled workflows needed to accelerate and improve the Incubator's screening and deep-dive process, and that process is written down and in use across the team.

You Will:

• Own the case for (or against) potential Big Bets: an opportunity to use AI to improve the lives of millions of people in LMICs
• Own a portfolio of program due diligence end to end: take each AI-enabled intervention you own from long list to rapid screen to deep dive, and be accountable for the quality and speed of the assessment you recommend to leadership.
• Drive the thinking, not just the research: frame the decision-relevant question for each program, find the crux quickly, and decide what evidence would change the answer.
• Build the analysis that underpins each decision: evidence reviews, back-of-the-envelope (BOTEC) and full cost-effectiveness estimates, risk assessments, landscape scans, expert interviews, and early operational and technical scoping.
• Make the recommendation: write clear, calibrated memos to 2AI leadership on whether an idea should advance to pilot, pause, or be set aside, and say how confident you are and why.
• Move winners from paper to pilot
• Co-design pilots that answer the decision-relevant question: working with 2AI leadership and program teams, define scope, sample, measurement, cost tracking, timeline, and success and stop criteria in advance. Favor fast, lean designs that answer the core question over comprehensive ones.
• Do the on-the-ground work: travel to prospective pilot locations to assess partners and delivery channels, negotiate scopes with implementers, and work in-country to stand up pilot infrastructure.
• Build the repeatable stack: data pipelines, monitoring dashboards, evaluation templates, and partner MOUs, so the second and third pilot designs are more efficient than the first.
• Help build the Incubator itself
• Co-design the process: work with 2AI leadership to refine the screening criteria, define key activities and sequencing, clarify roles between teams, and build the supporting tools. Continually refine the process, based on what you learn from real cases.
• Make the Incubator AI-native: show how a small research team uses frontier AI tools (Claude Code, agentic pipelines, LLM-assisted literature synthesis) to work well beyond its size. Build, test, and document the AI workflows behind each key step in the Incubator process.
• Support recruitment: help source, assess, and onboard future Incubator team members as the team grows.

Requirements

• The ideal candidate has a track record of structured, rigorous problem solving in fast, project-based environments: taking in a lot of open-ended information, structuring a path to answer a specific question, and getting to a clear recommendation for stakeholders.
• You have also used AI to meaningfully enhance your work, and are energized by the prospect of working to translate AI capabilities to practical applications. The role sits at the nexus of research, analytics, strategy and operations: it is for someone who wants to own the question "what should 2AI do next, and how do we know," and who enjoys being accountable for the answer.

You will likely have:

• 3-4+ years of experience in management consulting, a rigorous startup environment, or a high-caliber NGO, including time leading a workstream from an open question to a recommendation.
• Deep, hands-on experience with modeling and quantitative analysis: building models from scratch, including back-of-the-envelope estimates under uncertainty, and pressure-testing other people's assumptions. Experience with cost-effectiveness analysis and its benchmarks (GiveWell, Coefficient Giving, DCP3, WHO-CHOICE, GiveDirectly cash as a bar) is a strong plus, but any rigorous modeling or quantitative work counts.
• The ability to read research critically: telling a strong randomized controlled trial (RCT) from a weak observational study, and knowing what a given piece of evidence can and cannot support. Experience with research methods and evidence reviews is a strong plus.
• Experience engaging with technical experts, in fields where you yourself are not an expert, to gather key information and insights.
• Experience working in a fast-paced, project-based environment with short deliverable cycles, sprints, and frequent stakeholder check-ins.
• The ability to hone in on the crux of an argument in a field you do not know well and to communicate confidence levels legibly in writing.
• Comfort with ambiguity, high throughput, and the pace of an early-stage organization.
• Willingness to travel internationally (up to 25%) to pilot sites and partners.
• Heavy, daily use of frontier AI tools (Claude, Claude Code, or equivalent) in your analytical work, and a track record of pushing how you use them further.
• Excellent written communication, especially short decision memos that lead with the recommendation and are honest about uncertainty.
• Nice to have: Experience working on digital product uptake and integration (product strategy, user needs, adoption, unit economics), ideally including AI products, and ideally in an LMIC context, in either the for-profit or non-profit sector.
• Experience working in LMIC contexts and/or working in the international development sector.
• Research or monitoring, learning, and evaluation (M&E) experience, or NGO program experience.
• Domain depth in a sector 2AI has not yet entered (education, financial inclusion, WASH, climate adaptation).
• Position Location

This role location is flexible anywhere in North America for fully remote candidates. This role requires the ability to travel internationally up to 25% of the time. We are unable to sponsor or take over sponsorship of a U.S. employment visa at this time. Applicants must be legally authorized to work in the U.S. for roles based in the U.S.

Our process:

• Screening call: 30 mins - 1 hour
• Performance task: 2 hours (practical exercise)
• Team interviews: Two one-hour sessions with the people you'll work with most directly, including a short verbal case
• Final interview: 1 hour
• Reference checks

Use of AI in our hiring process: We believe strongly in the importance of human judgment throughout hiring, from screening to final decisions. We use AI to support our team across the process, including application review, note-taking, scheduling, and project management. AI does not make hiring decisions or reject candidates. We do not use models that train on inputs or candidate data.