🤖 Member of Technical Staff — AI Product & Developer Experience, Amazon AGI
I make AI products successful by making them adoptable. I own the product adoption strategy for Amazon's next-generation AI, from market positioning through activation and retention, working across Research, Product, Engineering, and GTM. I define adoption funnels, run the data to find where users drop off, and drive the product changes that fix it.
I think like a PM, execute like a builder, and measure like a growth team. I build with AI every day to help shape where software engineering is heading and keep my product instincts sharp.
product-coach: a review layer for product decisions. Coaching skills object to your plan using your own numbers. Then a scoreboard checks, weeks later, whether the objection was right. Advice that keeps score on itself.
bakeoff: pick a model by cost per correct call, not by leaderboard. Build an eval from your own agent logs, run it against any model on OpenRouter, and see accuracy, latency and cost side by side.
research: runnable notes on inference systems. Kernels, schedulers, profilers, and what it actually costs to serve models on NVIDIA, AMD, TPU, Trainium, Cerebras and d-Matrix silicon.
In 2026 I completed eight Product School certifications. Each one ended in a capstone: one product call, argued with data and shipped as a public repo.
| Certification | The problem | The call | Links |
|---|---|---|---|
| Product Leadership | A mental-wellness app loses users because it worked | Become an AI companion that builds memory over time. Never optimize for dependency | repo · deck |
| Product Analytics & Experimentation | Trial conversion stuck at 2% for 13 months | Fix activation, not acquisition. Test the leading metric first, because conversion alone would take 4.6 months to power | repo |
| AI Product Management | A P0 Slack thread nobody can triage fast enough | An AI copilot that turns the thread into a ranked, source-cited shortlist a PM can defend | repo · prototype |
| AI Product Strategy for Leaders | Product advice is never checked against what happened next | product-coach, as a real strategy: data flywheel, kill switch, pricing and board pitch | repo · deck |
| Claude Code for PMs | Four reliable responders went quiet after a release | Read interviews, tickets, metrics and routing code together, then prototype the fix | repo · report |
| Go-to-Market | Friends plan group trips in a chat, split costs over Venmo, and often never book | Every friend pays their own seat while the host holds the spots. Launch city by city | repo · deck |
| Agentic Workflows & Loops | Every PM writes the weekly leadership update by hand | An agent drafts it, an independent critic checks it, and it stops for a human. Autonomy is earned, not granted | repo · pitch |
| AI Evals | Should this competitive-intelligence agent ship? | Hold. A three-layer eval suite, judge calibration and CI gates show why, and what would change the answer | repo · deck |
📖 O'Reilly co-author: Generative AI on AWS (2023) and Data Science on AWS (2021)
🎓 DeepLearning.AI instructor and curriculum developer: Generative AI with Large Language Models (440,000+ learners) and Practical Data Science (45,000+ learners)
🎤 Keynotes and conference chairs: AI Engineer World's Fair, O'Reilly AI Superstream, Open Source Summit NA, MCP Dev Summit
👩🏼💻 Co-founder of the Düsseldorf chapter of Women in Big Data
📍 San Francisco Bay Area · antje.dev · LinkedIn · X · YouTube · AI Performance Engineering Meetup · AWS Blog