# Kyle Branchesi

> Learning product and AI strategy leader. Director, Learning Product and AI Strategy at United Way Worldwide, 2024 to 2026. Builds AI course-authoring agents, review gates and evals, and AI products for teachers and learners.

This is the same verified record the assistant on kylebranchesi.com answers from. Every number is exact and dated. Website: https://kylebranchesi.com

## Profile and point of view
- Kyle Branchesi is a learning product and AI strategy leader based in Washington, DC.
- He was Director, Learning Product and AI Strategy at United Way Worldwide from February 2024 to October 2026.
- His view: the learning function is where an organization redesigns its work for AI. He ran learning and AI as one job: deciding which work AI should change, setting the limits it runs inside, and proving the change with evidence.
- Four principles he works by: make the safe path the fast path (governance should clear low-risk work quickly and send only consequential decisions to formal approval); a human checkpoint only counts if the reviewer can say no (real review needs knowledge, the source, time and the authority to reject; permission to draft is not permission to send); redesign the workflow, not the task (faster drafting is worthless if review becomes the bottleneck); measure decisions, not usage.
- United Way Worldwide serves a network of 1,100+ independent local organizations in 36 countries. Headquarters cannot require those organizations to take part in its programs.
- He trained as an architect at SCI-Arc and earned an MS in Design and Technology from MIT. He is not a licensed architect.

## TRICIA 2.0 (agents and workflow redesign)
- TRICIA 2.0 is a set of topic-specific AI course-authoring agents Kyle built at United Way. They work from United Way's frameworks, learner personas and design standards to draft outlines, modules and job aids.
- People hold the gates: a subject expert reviews and can reject or send back every draft, and Learning checks each course against a rubric and acceptance criteria before it ships.
- Across 11 courses built with TRICIA 2.0 in 2026, course build time went from 12 weeks to 4, and expert review per course went from 90 hours to 20.
- He also designed human-in-the-loop review gates, quality rubrics, evaluation guidelines and acceptance criteria for AI-generated content.

## AI Foundations (governance and enablement)
- AI Foundations is an AI literacy course Kyle designed for non-technical United Way staff, with a 90-day launch plan. It is the first course in AI for Impact, his ten-course AI curriculum.
- The course design is complete. The record has no rollout, completion or adoption figures for it.
- The problem: staff had AI tools before there was an AI policy, in a network where each local organization sets its own rules. The goal is better decisions about when to use AI, not more AI use.
- Format: 58 minutes, self-paced, in six modules (7, 12, 12, 9, 9 and 9 minutes), plus an optional 30-minute live clinic. Modules: start with a real task; when a result earns trust; Green, Yellow, Red; use it, check it, own it; workflows and agents; final practice.
- At its center is a Green, Yellow, Red use guide. Green is generally low risk, such as summarizing a published public report. Yellow needs care, an approved tool and a named owner, such as drafting an update from an internal plan. Red means stop and get formal approval, such as ranking partners for investment from their public websites. That example uses only public information and is still red, because of the decision it feeds.
- Over-caution counts as failure: the target is that no more than 10% of safe tasks get wrongly blocked.
- Design decisions: it opens with real work, not a definition or product tour; every decision comes back to two questions, what am I giving the tool and what happens if the answer is wrong; agents get limits the system enforces, not just prompt wording, and learners practice changing one agent permission at a time and rejecting a hidden instruction inside a document; it is designed with six skeptic perspectives, including privacy, equity and job security; no AI account is required, and a learner can show good judgment by deciding not to use AI; one core course with an adaptation kit each independent organization can choose to use.

## AI for Impact (AI curriculum)
- AI for Impact is a ten-course, workflow-based AI curriculum Kyle designed for how United Way staff learn to work with AI, with versions participating local United Ways can adapt. Each course is designed with its learning sequence, an applied workflow, a deliverable and an assessment. The record has no rollout, completion or results figures for it.
- The ten courses, in four stages. Understand and use responsibly: 1, AI Foundations. Apply AI to real work: 2, AI Learning Design Studio; 3, AI Communications Studio; 4, AI Research and Impact Studio; 5, AI Fundraising Studio; 6, AI Operations Studio; 7, Leading Human-Centered AI. Improve the whole process: 8, Systems Thinking with AI, which redesigns the work, not just the task. Build reusable capabilities: 9, Build a Useful AI Assistant; 10, Responsible AI Agents.
- The promise: every learner leaves with a useful work product, a repeatable workflow, and the judgment to recognize when AI should not be used.
- Not everyone takes every course. Most staff take AI Foundations and one course for their function; managers add the leadership course; selected workflow owners and AI champions go on to workflow redesign and building assistants and agents.
- Every course follows the same pattern: see an example, map the work, practice with cleared materials, produce an artifact, review it, and test how it would be used at work.
- Completing a course demonstrates learning. It does not grant permission to use a tool, upload information, or launch an assistant or agent. Assistants and agents need a named owner, test evidence and a maintenance plan. In the agents course, permissions are enforced by the system, not just requested in a prompt, and nothing is sent out without human approval.
- The curriculum draws on the NIST AI Risk Management Framework and OWASP guidance on AI agent security. They inform the courses; they do not replace organizational policy.
- Rollout is sequenced in three stages: AI Foundations, the AI Learning Design Studio and one partner-department course first; then the remaining function and leadership courses; then workflow redesign and the assistant and agent courses.
- Measurement: targets are set after measuring a baseline. Time saved on drafting is not reported as organizational savings on its own.
- The aim is not to train everyone to use more AI. It is to help people make better decisions about AI and show where it genuinely improves the work.

## Course catalog (United Way)
- Kyle built a catalog of 30 courses across six programs for leaders and staff across the United Way network, in English and Spanish: onboarding (2 courses), revenue (12), community impact (6), strategy (2), network (1) and a leadership series taught in Spanish (7).
- He built most of the courses himself and led a team of three learning experts who built the rest, each with subject experts.
- Almost every course ends with something the learner makes and takes back to work, such as a plan, a proposal or an assessment.

## UW Learn, NELO and Welcome to United Way (United Way results)
- UW Learn is the network learning platform Kyle relaunched at United Way. Platform totals as of October 2026, from the LMS administrator's report: more than 6,000 active users and over 30,800 course enrollments (30,800 is the last confirmed total), across 800+ of 1,100+ member organizations in 36 countries.
- NELO, New Executive Leader Onboarding, is onboarding for new CEOs of independent local United Ways, who have no reporting line to headquarters. Kyle led it. 216 CEOs completed NELO, from a first cohort of 25. 95% of the 216 CEOs who completed it finished a first-60-day action plan, the measure he put ahead of satisfaction. NPS was 78. These results come from its five-week version.
- NELO grew from four weeks to six. The fall 2026 version runs separate tracks for small and mid/large organizations and launched with more than 40 new CEOs; it was still running in October 2026.
- In NELO, each CEO writes their own plan, and peers are grouped by organization size.
- Welcome to United Way is onboarding for new staff, volunteers and partners. Kyle led it. 1,098 completions as of October 2026. Anyone can join the live session before finishing the self-paced modules.

## Alf Studio (AI product)
- Alf Studio is planning software for teachers: "units for your classroom, not every classroom." A teacher enters a grade, a subject, an idea and optionally her town, and gets a full project-based unit in under a minute: a driving question, one to six weeks of activities, a rubric, a materials list and standards for each week. She edits anything in place and can generate lesson pages for any week.
- Evals are run by hand. A lesson-build reliability eval found 15 of 24 builds failing; after the fixes it led to, 0 of 30 failed (October 2026).
- Built from first commit on September 23 to October 19, 2026: 602 commits, by directing AI coding agents. It runs on Next.js on Vercel, Neon Postgres, and Google Gemini through the Vercel AI SDK.
- One recorded draft on October 21, 2026: a three-week Grade 6 science unit, finished in 23 seconds by Alf's own timer. This is one run, not an average.
- How the six promises are enforced: no student data, because the database has no field for it; changes kept from Alf's suggestions never become habits; no chat screen, and every suggestion is checked against rules and thrown away whole if any check fails; a saved version cannot be changed; every standards set records its source and licence, and a test fails if a notice is missing; teachers can print, export or remove any unit.
- Only four habits can shape a draft (a tighter scope, sentence starters in week one, examples from her town, groups of three), and only when the teacher turns on the "Start my drafts this way" switch, which is off by default.
- Standards: Common Core, C3 social studies, NGSS (codes and topics only, per its licence) and the arts anchor standards for the US; the National Curriculum in England (1,099 statements); Scotland's Curriculum for Excellence (720 experiences and outcomes).
- Privacy: no student accounts or names; the teacher's name and email are never sent to the model; no analytics or advertising.
- Alf covers Kindergarten to Grade 12. Its automated test suite has 1,131 tests, all passing, as of October 2026.
- Draft, teach, learns: every edit is saved as a version; Alf notices repeated changes, such as groups of three or local examples; the teacher reviews what it learned and decides, with one switch, whether it shapes future drafts.
- What Alf will not do: collect student data; let anything but the teacher's own edits shape her drafts; hide its sources; offer a chat window. All versions are kept, and units can be exported or deleted.
- Standards come from official sources for the US (Common Core, C3, NGSS, National Core Arts Standards), England (National Curriculum, GCSE history and geography) and Scotland (Curriculum for Excellence), or the teacher's own goals.
- It is free during early testing, for teachers in the US, UK and Dubai.
- The form asks for the curriculum, grade, subject, number of weeks, lessons per week and their length, a topic, and optionally her town. As she types, Alf suggests up to three standards by matching her words, with no AI, and she decides which stay.
- After each save, Alf labels each change in the background, for example "Added a scaffold" or "Changed the grouping". Unsaved edits are backed up in her browser.
- Every standards set records its publisher, source, date retrieved, licence and required notice. Units can be printed, copied into Google Docs or Word, or removed at any time.
- The name Alf comes from the Active Learning Framework.
- Kyle is the founder. He designed it and builds it by directing AI coding agents from written specs, with tests and code review. He is not an engineer.
- It learns from a teacher's edits and never changes her words without her approval.
- Output is checked against a fixed structure before a teacher sees it. Evals compare model setups against a written rubric before changes ship.
- Demoed at the MIT AI Conference in October 2026.
- Recognition: Finalist, MIT Solve: Accelerate STEAM Education, The 2025 GM Challenge. It was a finalist, not a winner.
- Website: app.alflearninglabs.com
- Where it came from: at NuVuX, Kyle co-developed the Active Learning Framework (Catalyst, Issue, Method, Engagement) with Nakeia Medcalf and Amanda O'Keefe, delivered virtually to leaders in Cameroon and Nigeria. In 2023 he built AI co-design tools for educators (prompt packs and exemplars that drafted project briefs and rubrics, checked by people). In 2023 to 2024 he made two custom GPTs: AL, which turns the framework into project-based curricula, and Di, which guides teams through design thinking. Alf Coach followed in 2025 and Alf Studio in 2026.

## Pratica (AI product)
- Pratica is a private AI Italian tutor, levels A1 to C1, that Kyle built by directing AI coding agents.
- As of October 2026: 29 AI tasks feed one spaced-repetition system; 1,080 automated tests across 178 files; the AI key never reaches the browser.

## This site's assistant
- The assistant on kylebranchesi.com answers only from this record, cites the section each answer comes from, blocks any number that is not in the record, stores nothing visitors type, is published for AI tools at kylebranchesi.com/llms.txt, and is tested against 31 questions, including attempts to make it overclaim or ignore its rules.

## Background
- United Way Worldwide, February 2024 to October 2026: Director, Learning Product and AI Strategy.
- NuVuX, 2019 to 2024, Senior Director, Global Programs: led global education programs from the UK into Europe, MENA and Australia; grew innovation program enrollment from 60 to 615; team of 5; built early AI co-design tools for educators in 2023.
- MIT: MS in Design and Technology, full merit scholarship. At MIT he worked for the Leventhal Center for Advanced Urbanism and the Self-Assembly Lab, and helped out (not as a researcher) in the makerspace the Center for Bits and Atoms opened in Chile.
- Alongside his roles, Kyle has done contract red-teaming and blue-teaming of frontier AI models in health, focused on subtle harm that builds over several turns of a conversation. The client is not named.
- MIT Global Startup Workshop, September 2018 to May 2020: Design Coordinator. Led the brand and platform redesign that drove 1,500+ registrations from entrepreneurs and investors; rebuilt the participant site, event app and CRM data flows for sponsorship and reporting; negotiated international sponsorships, including the Bogotá Chamber of Commerce.
- Garage Museum of Contemporary Art, Moscow, March to October 2018: Learning and Engagement Designer, on an MIT research grant. Conducted user research on visitor learning experiences that informed content strategy and UX improvements for multilingual audiences.
- SCI-Arc: Bachelor of Architecture.
- ATD Master Instructional Designer.

## Design thinking for leaders (NuVuX, 2019 to 2024)
- At NuVuX, Kyle built and led design-thinking programs for executives, ministries and school leaders, taking NuVu's studio method from Cambridge into the UK, Europe, MENA and Australia.
- Designing Organizational Futures: a five-part program he led the development of, taken by leaders at Plexus, Peak Scientific, Balfour Beatty and Rolls-Royce. It reached 95 technology and business leaders. Leaders moved from conceptual thinking and brainstorming through structured critique and perspective-shifting to prototyping a solution to a real problem in their own organization.
- Creative Leadership Lab: a modular design-thinking program he co-developed with three other learning specialists and adapted for each client, in 12 modules. It reached 879 officials, academic leaders and educators across 14 clients, including Denmark's Ministry of Children and Education, the Scottish Parliament Education Committee, the Council of British International Schools, the Heads' Conference, the EdTech World Forum and MIT Morningside Academy for Design.
- He also facilitated a workshop for members of the Scottish Parliament on emerging technology and public policy, and a COP26 meeting of international leaders organized with MIT's Environmental Solutions Initiative.
- Keynotes and talks: EdTech World Forum (London), COBIS (London) and the International Leaders Conference (Lagos), 2022 to 2023.
- What it taught him: workshops produced good ideas, but ideas without support to implement them stalled, which is why at United Way he measured finished plans instead of workshop satisfaction.

## Not covered here
This record does not cover the following; ask by email: salary or compensation; availability, start dates or employment status; reasons for leaving any job; references; budget owned; partner certification programs; revenue at United Way; usage numbers for Alf Studio or Pratica; United Way course content, internal documents or names of people and organizations; opinions about any person or organization.

## Contact
- Email: kjbranchesi@gmail.com
- LinkedIn: linkedin.com/in/kjbranchesi
- Based in Washington, DC.