Phil Wagner

Phil Wagner

Innovation for everyone.

I build the thing, then teach a few million people to use it. From naval nuclear reactors to AI, and whatever comes next.

Work samples Get in touch

Start here

Six things I thought you would want first, because you came to the AI enablement page. See everything

AI

2024–2026 · LLMs to AI

calling a model and reading the answeragents doing the work, with humans deciding where review still has to happen

Prompting gave way to systems that plan and act on their own. The question stopped being what a model can do and became which work you can delegate, and where a human should be involved.

My part. Built the fluency and agentic engineering program that took tens of thousands of people from using AI to building products and features with it. Synthesized from research what good prompts look like and how to define evals. While training my team, I built documentation that maintains itself.

What it took

Rapidly building and refining resources to support experts and champions for technical and non-technical communities.

What we were up against

A spectrum of users: from people already AI fluent from nights and weekends who needed guardrails to those who had never touched the tools. Overnight, everyone was vibe coding, generating documents, slides, prototypes, and long unedited and sometimes unreviewed writing. The hype made it all look effortless, but what would become of our commitments to reliability, security, and speed?

LLM

2019–2024 · Bespoke ML to LLMs

a purpose-built model for every problemone general model, prompted, and a new question of whether to trust it

The economics inverted. Instead of gathering a dataset and training a model per task, teams reached for a general model and tuned it with language, which shifted the problem from training to evaluation and trust.

My part. Wrote the guide that turned prompting from folklore into documented practice, and built the course that gave people hands-on time with the models.

What it took

Nobody agreed what made a prompt good. The models were weak, so some were stuffing everything into the context while others swore they'd never use them. The guide had to be synthesized from experiments rather than opinion, and running a company-wide LLM bootcamp gave daily signals on how the models behaved and for employees, their levels of trust, and productivity. In order to develop a product development training, we interviewed dozens of product managers inside the core products and within DeepMind. That research became Building with Gemini and a company-wide community of practice.

What we were up against

ML was everywhere in the company and all those bespoke models were expensive and hard to improve. Then the Transformer paper changed everything. Suddenly we were experiencing a seismic shift we hadn't experienced since shifting to mobile platforms.

ML

2016–2019 · Statistics and heuristics to ML

hand-tuned heuristics and classical statisticsmachine learning is a core facet of engineering

Machine learning stopped being a research specialty and became something an ordinary product engineer was expected to reach for. Since it affected all areas from data collection to deploying, everyone needed to learn it ASAP.

My part. Tech Lead for the course that taught ML to millions from interactivity to internationalization. Adapted the content for non-engineers to enable them to participate in the discussion and launched the initial AI Principles research site. Designed an experience for large enterprises.

What it took

PhD-level equations and concepts needed to be translated into hands-on low/no-code experiences so engineers, product teams, and lawmakers alike can play with linear regression, embeddings, and data engineering. Our global audience required an internationalization pipeline, but most of the technical words in ML had no counterpart. We developed high-touch experiences for startups and large enterprise customers.

What we were up against

The people who knew how to build models were PhD one-of-a-kind researchers. Our content had to support engineers who had learned theory during the AI winter to people who stopped at algebra. Universities and MOOCs already had the theory covered, so we wanted ours to be full of the practical lessons we learned along the way. When I joined the project, the SMEs had built a 400 slide deck, requiring a week of training, and hours of cloud setup.

Mobile

2011–2019 · Web to Mobile

web applications on a computernative mobile applications and design for everyone

The smartphone broke all assumptions for software engineering. No guaranteed high-quality connection, no mouse, no large screen, no user with all five senses, and a new attack surface. Qualities that had been someone else's problem became conditions of shipping.

My part. Integrated the required mindset shift and skills into all existing trainings. Shifting the implications for privacy, security, reliability and accessibility from policy statements into engineering practice.

What it took

The training had to do two things at once: make the case that none of these could be traded away before shipping, and be practical enough to implement in a real product.

What we were up against

Product teams are measured on market fit and user delight, and everything else competes for attention. Regulation, advocacy groups and new markets moved those priorities for us. So did attackers phishing credentials, and governments with an interest in a less open web.

CS

2010–2014 · CS for everyone

computer science as a separate elective for a fewcomputational thinking in every subject, for everyone

Computing was a vocational elective outside the core curriculum, often little more than office tools and typing. The goal was to integrate algorithmic thinking into math, science and the humanities, not only in front of future CS majors.

My part. Wrote the lessons that put it inside existing subjects aligned to core standards and trained the teachers who had no CS department behind them. Built the first Hour of Code activity.

What it took

I developed the design principles and we codesigned with students and research partners. I socialized my own efforts first through my blog, sharing how math and science teachers were applying these principles, and later they were built into the lessons I made for Google, into conference presentations, and into professional development. Students were not only understanding computational thinking concepts, they were playing with them, as young as five.

What we were up against

Software was growing faster than computer science could supply engineers, and fresh ideas were more likely to come from people passionate about other subjects who learned to code. The constraint was that teachers were already buried in standards-based curriculum. Nothing we added could pull them away from it.

ALL

Running underneath it all

STEM

2000–2011 · STEM Classroom Educator

Background

Recognition

California Teacher of the Year

Air Force Association

Awarded by the Air Force Association for excellence in secondary STEM instruction and cultivating an inquiry-driven computer science environment at High Tech High.

Education

  • MA, Educational Technology and Instructional Design San Diego State University
  • BS, Physics and Secondary Education Colorado Christian University
  • Teaching Credential, Secondary Math and Science California

Speaking and community

  • ISTE + ATD International Society for Technology in Education
  • SIGCSE Special Interest Group on Computer Science Education
  • CSTA Computer Science Teachers Association
  • AAPT American Association of Physics Teachers
  • Google for Education and Developers
  • NSTA Panelist

Partnerships

  • DesignScience
  • Raspberry Pi
  • Arduino
  • Scratch

Competencies

Every capability carries the shifts it was proven in. One I cannot point at a piece of work for is not listed, so nothing here is claimed without something behind it.

Choosing a capability filters the page to the work that proves it. Nothing here scores or ranks, marks are only ever added.

Learning & Instructional Design25

  • CSMobileML
  • Mobile
  • MLAI
  • STEMCSMobileML
  • CSML
  • CSMobileML
  • MLALL
  • CS
  • CSALL
  • MobileML
  • Mobile
  • CSMobile
  • STEMMobileML
  • CS
  • CSMobile
  • MLAI
  • MLAI
  • STEMMobile
  • MobileML
  • STEM
  • CSMobileMLALL
  • STEMMobile
  • MobileALL
  • ALL
  • CS

Measurement & Platforms7

  • MobileML
  • MobileMLAIALL
  • CSML
  • ML
  • CSML
  • CS
  • ML

AI12

  • MLLLMAI
  • LLMAI
  • AI
  • LLMAI
  • LLMAI
  • LLMAI
  • LLM
  • AI
  • MLLLM
  • AI
  • LLM
  • LLM

Technical Writing & DevEx21

  • CSMLLLM
  • CSMobile
  • MobileLLMAI
  • AI
  • MobileMLLLM
  • LLMAI
  • LLM
  • Mobile
  • AI
  • LLM
  • MobileAI
  • AI
  • MobileML
  • MobileLLM
  • STEMMobile
  • AI
  • AI
  • ML
  • CS
  • CSAI
  • ML

Technical9

  • CSMobileML
  • CSMobileMLAI
  • CSMobileMLAI
  • MobileMLAI
  • CSAI
  • CSML
  • CS
  • CSMobile
  • CS

Leadership & Program23

  • CSMobileMLAI
  • MobileML
  • ALL
  • ALL
  • AI
  • MobileAI
  • AIALL
  • MLLLMAI
  • MLAI
  • AIALL
  • AIALL
  • MobileML
  • AIALL
  • CSLLMALL
  • MobileML
  • MLAI
  • MLALL
  • CSALL
  • MLAI
  • ML
  • AI
  • Mobile
  • STEMMobileML

K-12 Classroom23

  • STEM
  • STEM
  • STEM
  • CS
  • CS
  • CS
  • STEM
  • CS
  • STEM
  • STEM
  • STEM
  • STEM
  • STEM
  • STEM
  • STEM
  • STEM
  • ALL
  • STEM
  • STEM
  • STEM
  • STEM
  • ALL
  • ALL

Say hello

Learning design, AI enablement, developer education, or just comparing notes on teaching.

or [email protected]